DETAILED ACTION
Claims 1-20 are presented for examination. Claims 1, 8, and 15-18 are currently amended. This office action is in response to the submission on 8/3/2026.
Response to Arguments
With Respect to the 35 USC § 102 and 103 Rejections:
Applicant’s arguments, see pages 12-19 of applicant response filed 8/3/2026, with respect to the rejections of claims 1, 8, and 15 under 35 USC § 102 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new grounds of rejection is made in view of Gupta, I., Samandarli et al. "Autoregressive and Machine Learning Driven Production Forecasting - Midland Basin Case Study."
Applicant’s arguments, see pages 19-22 of applicant response filed 8/3/2026, with respect to the rejections of claims 7 and 14 under 35 USC § 10have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new grounds of rejection is made in view of Gupta, I., Samandarli et al. "Autoregressive and Machine Learning Driven Production Forecasting - Midland Basin Case Study."
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-6, 8-13 and 15-20 are rejected under 35 U.S.C. 103 as being unpatentable over Devarkonda (US20220067622A1) in view of Gupta, I., Samandarli et al. "Autoregressive and Machine Learning Driven Production Forecasting - Midland Basin Case Study." Paper presented at the SPE/AAPG/SEG Unconventional Resources Technology Conference, Houston, Texas, USA, July 2021. doi: https://doi.org/10.15530/urtec-2021-5184 (hereinafter referred to as “Gupta”).
Claim 1:
Devarkonda teaches “A predictive control system for a plant comprising a subprocess and a main process affected by the subprocess” (Devarkonda teaches a final-assembly model 122 which predicts product-throughput i.e. it is a main process and that it is affected by first, second, and third component models i.e. subprocesses in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120." and Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."),
“the predictive control system comprising: a model generator configured to: obtain actual historical values of one or more forecast variables representing one or more outputs of the subprocess, actual historical values of one or more controlled variables representing one or more outputs of the main process, and historical values of one or more manipulated variables representing one or more inputs to the plant;” (Devarkonda teaches the embodiments disclosed involve prediction of production capacity using production data from multiple levels of a production stream i.e. historical values of outputs of subprocesses and main processes in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources."; Devarkonda teaches data stores that contain historical data in Devarkonda [0039] "Each of the data stores provides historical data relevant to the component produced by the corresponding factory. This historical data can include data items such as raw materials used, amount of material used i.e. inputs to the plant, manufacturing steps, configuration of the manufacturing steps, actual duration of the manufacturing steps, quality metrics, energy expense, labor expense, production capacity of equipment, time taken to transport raw materials or parts i.e. historical values of outputs of subprocesses and/or any one or more other data items deemed suitable by those of skill in the art for a given implementation.... In at least one embodiment, each of the first-component historical data 140, the second-component historical data 142, and the third-component historical data 144 includes historical data pertaining to the corresponding operational metrics of the corresponding factory. i.e. historical values of outputs of the main process" (emphasis added)),
“train a subprocess model representing the subprocess” (Devarkonda teaches that the models are trained using historical observations of production metrics in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources. Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans. Some embodiments involve inferring, using the machine-learning models and the observed production operational metrics, causal factors that impact the predicted production operational metrics. Some embodiments also involve generation of prognostic alerts, action recommendations based on simulation and generative models, and/or predicted value impacts for a plurality of actions related to the operation of the production value stream. As used herein, “causal factors” include those factors that are elements that contribute in some manner to (e.g., determine) the outcome of a process."),
“train a main process model representing the main process” (Devarkonda teaches a final-assembly model 122 which predicts product-throughput i.e. it represents a main process in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches that the models are trained using historical observations of production metrics and plans derived from the production value stream i.e. historical values of controlled variables in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources. Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans. Some embodiments involve inferring, using the machine-learning models and the observed production operational metrics, causal factors that impact the predicted production operational metrics. Some embodiments also involve generation of prognostic alerts, action recommendations based on simulation and generative models, and/or predicted value impacts for a plurality of actions related to the operation of the production value stream. As used herein, “causal factors” include those factors that are elements that contribute in some manner to (e.g., determine) the outcome of a process."),
“a controller configured to execute a predictive control process using the subprocess model and the main process model to control operation of the plant,” (Devarkonda teaches a machine which instructions that cause it to perform the methods discussed in Devarkonda [0133] "FIG. 16 is a diagrammatic representation of a machine 1600 within which instructions 1612 (e.g., software, a program, an application, an applet, an app, or other executable code) for causing the machine 1600 to perform any one or more of the methodologies discussed herein may be executed." and in Devarkonda [0134] "The machine 1600 may include processors 1602, memory 1604, and I/O components 1606, which may be configured to communicate with each other via a bus 1644."; Devarkonda teaches an action-and-alert process that provides recommended actions 168 to an implementation interface 160 in Devarkonda [0043] "The causal-analysis machine-learning model 124 provides identified causal factor 154 to an action-and-alert process 126. Based at least in part on the identified causal factor 154, the action-and-alert process 126 generates alerts 156 and recommended actions 168, and transmits both to the implementation interface 160, which could include one or more user interfaces for human users (e.g., graphical user interfaces (GUIs), audiovisual interfaces, and/or the like) and/or one or more automated interfaces for carrying out automated processing of the alerts 156 and recommended actions 168."; Devarkonda teaches that the implementation interface transmits commands to alter operating parameters i.e. control operation of the plant in Devarkonda [0046] "The implementation interface 160 generates implementation commands 162 and transmits the implementation commands 162 to the final-assembly/production process 108 in order to alter one or more operating parameters of the final-assembly process 108. In some embodiments, although not pictured in FIG. 1, the implementation interface 160 also or instead generates commands for transmission to one, some, or all of the upstream nodes (e.g., the first-component factory 102, the second-component factory 104, and/or the third-component factory 106), in order to alter one or more operating parameters of the operations there."),
“the predictive control process comprising: using the subprocess model to predict future values of the one or more forecast variables based on values of the one or more manipulated variables;” (Devarkonda teaches a first-component machine learning model i.e. subprocess model which provides first-component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150. Any machine-learning model described herein can be implemented using a machine-learning program executing on one or more hardware platforms, and can be operated as a standalone machine-learning model and/or be combined with one or more other machine-learning models in various different embodiments."; Devarkonda teaches that the models make predictions based on observed production metrics i.e. manipulated variables in Devarkonda [0035] "Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."),
“using the main process model to predict future values of the one or more controlled variables based on the future values of the one or more forecast variables predicted by the subprocess model;” (Devarkonda teaches the final-assembly model receives component predictions from the first component model in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches the final-assembly model generates product-level predictions in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."), and
“and controlling operation of the plant based on the future values of the one or more controlled variables predicted by the main process model.” (Devarkonda teaches a causal-analysis model that receives the predictions from the first-component model and final-assembly model and provides a causal factor to an action-and-alert process which generates recommended actions in Devarkonda [0043] "As described, the causal-analysis machine-learning model 124 also receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120. Based on all of those sets of predictions, the causal-analysis machine-learning model 124 infers causal factor of one or more of the predictions generated by one or more of the first-component machine-learning model 116, the second-component machine-learning model 118, the third-component machine-learning model 120, and the final-assembly machine-learning model 122. The causal-analysis machine-learning model 124 provides identified causal factor 154 to an action-and-alert process 126. Based at least in part on the identified causal factor 154, the action-and-alert process 126 generates alerts 156 and recommended actions 168, and transmits both to the implementation interface 160"; Devarkonda teaches the implementation interface 160 may alter operating parameters of the process in Devarkonda [0046] "The implementation interface 160 generates implementation commands 162 and transmits the implementation commands 162 to the final-assembly/production process 108 in order to alter one or more operating parameters of the final-assembly process 108.").
Devarkonda does not appear to explicitly teach “train a subprocess model representing the subprocess, wherein training the subprocess model comprises (i) using the subprocess model to generate predicted historical values of the one or more forecast variables as a function of prior predicted historical values of the one or more forecast variables generated by the subprocess model and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more forecast variables generated by the subprocess model with the actual historical values of the one or more forecast variables;” or “train a main process model representing the main process, wherein training the main process model comprises (i) using the main process model to generate predicted historical values of the one or more controlled variables as a function of prior predicted historical values of the one or more controlled variables generated by the main process model, the actual historical values of the one or more forecast variables, and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more controlled variables generated by the main process model with the actual historical values of the one or more controlled variables;” however, Gupta does teach these claim limitation.
Gupta teaches “train a subprocess model representing the subprocess, wherein training the subprocess model comprises (i) using the subprocess model to generate predicted historical values of the one or more forecast variables as a function of prior predicted historical values of the one or more forecast variables generated by the subprocess model and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more forecast variables generated by the subprocess model with the actual historical values of the one or more forecast variables;” (Gupta teaches a model where a predicted output is provided back to the model as input in order to predict the production of a component in a next time step in addition to prior production data i.e. historical values of controlled variables in Gupta [Page 5, first two paragraphs] "The ML-based forecasts were generated from an autoregressive ensemble model. This model accepts as inputs well completion parameters, subsurface geology, inter-well spacing, and historical production, when available. Inputs are subject to standard preprocessing (e.g. encoding any categorical data as one hot vectors), and missing values are imputed, but there are no outlier removal or correction steps. Feature selection is used to choose the n best performing features, where n is determined empirically via tuning. The final regressor used was an extremely randomized trees (extra-trees or ET) model, which was chosen in part for its robust handling of noise. The model output is a per-well timeseries of actual and forecasted production, indexed by IP day (days since well inception). For a given well, the model starts forecasting at the first timepoint t, using the production history from a fixed-size window into the past, t-w:t, and predicting values up to a fixed-size horizon, t:t+h. The model then moves h steps forward in time, and uses its own prior predictions along with older production history t+h-w:t+h to generate a forecast over the next horizon, t+h:t+2h."; Gupta teaches the model is tested by holding back data during training and comparing the predicted values to the actual values in Gupta [Page 6, Evaluation Metrics section, first paragraph] "Standard benchmarks were set for machine learning forecasts which must be met before machine learning can be integrated in traditional forecasting processes. 80% of the total well set was taken for training the models and the remaining 20% was withheld for testing the trained models. A series of chronological back tests, also called time machine testing, were used to evaluate the short-term and long-term performance of the machine learning models. For time machine testing, production data for all the training wells for up to 2 years from the current production date was held back i.e. not exposed to the machine learning models for training. The model predictions were then compared with this held back production to test the model performance."), and
“train a main process model representing the main process, wherein training the main process model comprises (i) using the main process model to generate predicted historical values of the one or more controlled variables as a function of prior predicted historical values of the one or more controlled variables generated by the main process model, the actual historical values of the one or more forecast variables, and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more controlled variables generated by the main process model with the actual historical values of the one or more controlled variables;” (Gupta teaches a model where a predicted output is provided back to the model as input in order to predict the production of a component in a next time step in addition to prior production data i.e. historical values of controlled variables in Gupta [Page 5, first two paragraphs] "The ML-based forecasts were generated from an autoregressive ensemble model. This model accepts as inputs well completion parameters, subsurface geology, inter-well spacing, and historical production, when available. Inputs are subject to standard preprocessing (e.g. encoding any categorical data as one hot vectors), and missing values are imputed, but there are no outlier removal or correction steps. Feature selection is used to choose the n best performing features, where n is determined empirically via tuning. The final regressor used was an extremely randomized trees (extra-trees or ET) model, which was chosen in part for its robust handling of noise. The model output is a per-well timeseries of actual and forecasted production, indexed by IP day (days since well inception). For a given well, the model starts forecasting at the first timepoint t, using the production history from a fixed-size window into the past, t-w:t, and predicting values up to a fixed-size horizon, t:t+h. The model then moves h steps forward in time, and uses its own prior predictions along with older production history t+h-w:t+h to generate a forecast over the next horizon, t+h:t+2h."; Gupta teaches the model is tested by holding back data during training and comparing the predicted values to the actual values in Gupta [Page 6, Evaluation Metrics section, first paragraph] "Standard benchmarks were set for machine learning forecasts which must be met before machine learning can be integrated in traditional forecasting processes. 80% of the total well set was taken for training the models and the remaining 20% was withheld for testing the trained models. A series of chronological back tests, also called time machine testing, were used to evaluate the short-term and long-term performance of the machine learning models. For time machine testing, production data for all the training wells for up to 2 years from the current production date was held back i.e. not exposed to the machine learning models for training. The model predictions were then compared with this held back production to test the model performance."; The concept of training the model using its own predictions as inputs may be applied to the model of Devarkonda which a person having ordinary skill in the art would recognize as an improvement to the model.).
Devarkonda and Gupta are analogous art because they are from the same field of endeavor of making predictions using models. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having teachings of Devarkonda and Gupta before him/her, to modify the teachings of Systems and methods for automating production intelligence across value streams using interconnected machine-learning models of Devarkonda to include the training and validation of a model by using both known inputs and previously predicted outputs as inputs to a model of Gupta because adding the Autoregressive and Machine Learning Driven Production Forecasting of Gupta would allow for improved forecasting as described in Gupta [Pages 11-12, first 2 paragraphs of “Discussion” section] “This successfully demonstrated the potential of ML-based approach as a practical replacement for Arps based forecasting technique. Although the primary objective of the pilot was to test the accuracy of the ML-based techniques, we strongly believe the biggest value will come from workflow automation and efficiency. In a market where efficiencies are crucial for the organization’s bottom-line, our mission is to achieve the best results. In the pre-drill models, the largest relative errors occur during early time for the oil phase, which may result from operating conditions for a well such as tank battery constraints, artificial lift selection, and artificial lift intensity. Over time, pre-drill relative errors stabilized at 2-4% range and became asymptotic. Emphasis was placed on improving oil phase forecasts as the primary driver for economic value during the life of the well; therefore, oil phase models yield better results than the other two phases. In the future, additional work will be performed to elevate the accuracy of gas and water phases. In the post-drill models, satisfactory accuracy levels were also obtained. During two years of production, the program level error stayed within +/- 5%, which was better than traditional Arps for the same well set. This outperformance can be attributed in part to ML-based approach’s flexibility to not follow a defined equation.”
Claim 2:
Devarkonda in view of Gupta teaches “The predictive control system of claim 1, wherein:
the model generator is configured to independently train the main process model without using predictions generated by the subprocess model;” (Devarkonda teaches that the models may be trained using supervised learning and ran against a training dataset in Devarkonda [0108-0109] "During a learning phase, the models are developed against a training dataset of inputs to optimize the models to correctly predict the output for a given input. Generally, the learning phase may be supervised, semi-supervised, or unsupervised; indicating a decreasing level to which the “correct” outputs are provided in correspondence to the training inputs. In a supervised learning phase, all of the outputs are provided to the model and the model is directed to develop a general rule or algorithm that maps the input to the output... Models may be run against a training dataset for several epochs (e.g., iterations), in which the training dataset is repeatedly fed into the model to refine its results. For example, in a supervised learning phase, a model is developed to predict the output for a given set of inputs, and is evaluated over several epochs to more reliably provide the output that is specified as corresponding to the given input for the greatest number of inputs for the training dataset."; Devarkonda teaches that the models are trained using historical observations of production metrics i.e. it doesn't use predictions generated by the subprocess model in Devarkonda [0035] "The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."), and
“and the controller is configured to use the predictions generated by the subprocess model as inputs to the main process model when executing the predictive control process.” (Devarkonda teaches the final-assembly model i.e. main process model receives component predictions from the first component model i.e. subprocess model in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120.").
Claim 3:
Devarkonda in view of Gupta teaches “The predictive control system of claim 1, wherein controlling operation of the plant comprises: evaluating a reward function using the future values of the one or more controlled variables predicted by the main process model; adjusting the values of the one or more manipulated variables to drive the reward function toward an extremum; and using the adjusted values of the one or more manipulated variables to control operation of the plant.” (Devarkonda teaches that recommended actions are prioritized based on minimizing a cost function or maximizing a reward function and that implementation commands may be transmitted to the production process to alter operating parameters in Devarkonda [0045-0046] "Moreover, the recommended actions 168 in some embodiments are prioritized based on quantified risk (e.g., based on minimizing a cost function, maximizing a reward function, and/or the like, where such function could take into account upstream materials, parts, components, process steps, and/or the like). The quantified risk can be based on operational measurements across the production value stream, relating to, e.g., upstream material deliveries, modules, process subsystems, operator efficiency, etc. Action-sequence recommendations can be generated at least in part by minimizing a cost function that incorporates operational and financial metrics such as revenue, profits, operating margin, lost sales, penalties due to late deliveries, expedite costs, impact on new contracts due to non-compliance, inventory-holding costs, fill rates, production quantity, and/or the like... The implementation interface 160 generates implementation commands 162 and transmits the implementation commands 162 to the final-assembly/production process 108 in order to alter one or more operating parameters of the final-assembly process 108.”).
Claim 4:
Devarkonda in view of Gupta teaches “The predictive control system of claim 1, wherein: the plant comprises a plurality of subprocesses and a plurality of main processes affected by the plurality of subprocesses;” (Devarkonda teaches that a final-assembly model i.e. main process model receives predictions from multiple component models i.e. subprocess models in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches a causal-analysis model i.e. a second main process model in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."; Devarkonda teaches that models may be deployed in connection with raw-material providers, sub-assembly providers, and others i.e. there may be models providing information to the component models and they may be interpreted as main process models in Devarkonda [0040] "The three example upstream machine-learning models that are depicted in FIG. 1 are examples of upstream machine-learning models that correspond to providers of components that are later combined into an end product. This is one example of a type of upstream machine-learning model that can be deployed in embodiments of the present disclosure. Other examples include machine-learning models deployed in connection with raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like. Other examples include machine-learning models deployed in connection with process manufacturing raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like."),
“the model generator is configured to train a plurality of subprocess models representing the plurality of subprocesses and a plurality of main process models representing the plurality of main processes;” (Devarkonda teaches that the models may be trained using supervised learning and ran against a training dataset in Devarkonda [0108-0109] "During a learning phase, the models are developed against a training dataset of inputs to optimize the models to correctly predict the output for a given input. Generally, the learning phase may be supervised, semi-supervised, or unsupervised; indicating a decreasing level to which the “correct” outputs are provided in correspondence to the training inputs. In a supervised learning phase, all of the outputs are provided to the model and the model is directed to develop a general rule or algorithm that maps the input to the output... Models may be run against a training dataset for several epochs (e.g., iterations), in which the training dataset is repeatedly fed into the model to refine its results. For example, in a supervised learning phase, a model is developed to predict the output for a given set of inputs, and is evaluated over several epochs to more reliably provide the output that is specified as corresponding to the given input for the greatest number of inputs for the training dataset."; Devarkonda teaches that the models are trained using historical observations of production metrics in Devarkonda [0035] "The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."), and
“and the controller is configured to use the plurality of subprocess models to predict future values of the one or more forecast variables representing outputs of the plurality of subprocesses and use the plurality of main process models to predict future values of the one or more controlled variables representing outputs of the plurality of main processes.” (Devarkonda teaches component models providing component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150"; Devarkonda teaches a final-assembly model providing product predictions in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."; Devarkonda teaches the causal-analysis model 124 providing causal factors in Devarkonda [0043] "The causal-analysis machine-learning model 124 provides identified causal factor 154 to an action-and-alert process 126. Based at least in part on the identified causal factor 154, the action-and-alert process 126 generates alerts 156 and recommended actions 168, and transmits both to the implementation interface 160, which could include one or more user interfaces for human users (e.g., graphical user interfaces (GUIs), audiovisual interfaces, and/or the like) and/or one or more automated interfaces for carrying out automated processing of the alerts 156 and recommended actions 168."; Devarkonda teaches that models may be deployed in connection with raw-material providers, sub-assembly providers, and others i.e. there may be models providing information to the component models and they may be interpreted as main process models in Devarkonda [0040] "The three example upstream machine-learning models that are depicted in FIG. 1 are examples of upstream machine-learning models that correspond to providers of components that are later combined into an end product. This is one example of a type of upstream machine-learning model that can be deployed in embodiments of the present disclosure. Other examples include machine-learning models deployed in connection with raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like. Other examples include machine-learning models deployed in connection with process manufacturing raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like.").
Claim 5:
Devarkonda in view of Gupta teaches “The predictive control system of claim 4, wherein the plurality of subprocesses are arranged in series with each other such that an output of a first subprocess of the plurality of subprocesses is provided as an input to a second subprocess of the plurality of subprocesses, wherein the controller is configured to: use a first subprocess model of the plurality of subprocess models representing the first subprocess to predict future values of one or more first forecast variables representing one or more outputs of the first subprocess based on the values of the one or more manipulated variables; use a second subprocess model of the plurality of subprocess models representing the second subprocess to predict future values of one or more second forecast variables representing one or more outputs of the second subprocess based on the future values of the one or more first forecast variables predicted by the first subprocess model;” (Devarkonda teaches component models providing component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150"; Devarkonda teaches a final-assembly model providing product predictions i.e. the final-assembly model may be regarded as a subassembly model in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."; Devarkonda teaches that models may be deployed in connection with raw-material providers, sub-assembly providers, and others i.e. there may be models providing information to the component models in Devarkonda [0040] "The three example upstream machine-learning models that are depicted in FIG. 1 are examples of upstream machine-learning models that correspond to providers of components that are later combined into an end product. This is one example of a type of upstream machine-learning model that can be deployed in embodiments of the present disclosure. Other examples include machine-learning models deployed in connection with raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like. Other examples include machine-learning models deployed in connection with process manufacturing raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like."),
“and use the main process model to predict future values of the one or more controlled variables based on the future values of the one or more second forecast variables predicted by the second subprocess model.” (Devarkonda teaches a final-assembly model providing product predictions i.e. the main process model predicts future values in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."; Devarkonda teaches that models may be deployed in connection with raw-material providers, sub-assembly providers, and others i.e. there may be models providing information to the component models in Devarkonda [0040] "The three example upstream machine-learning models that are depicted in FIG. 1 are examples of upstream machine-learning models that correspond to providers of components that are later combined into an end product. This is one example of a type of upstream machine-learning model that can be deployed in embodiments of the present disclosure. Other examples include machine-learning models deployed in connection with raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like. Other examples include machine-learning models deployed in connection with process manufacturing raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like.").
Claim 6:
Devarkonda in view of Gupta teaches “The predictive control system of claim 4, wherein the plurality of subprocesses are arranged in parallel with each other such that a first output of a first subprocess of the plurality of subprocesses and a second output of a second subprocess of the plurality of subprocesses are provided as inputs to the main process,” (Devarkonda teaches component models providing component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150"; Devarkonda teaches a final-assembly model providing product predictions based on the predictions of the component models in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120. " and in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."),
“wherein the controller is configured to: use a first subprocess model of the plurality of subprocess models representing the first subprocess to predict future values of one or more first forecast variables representing one or more outputs of the first subprocess based on the values of the one or more manipulated variables; use a second subprocess model of the plurality of subprocess models representing the second subprocess to predict future values of one or more second forecast variables representing one or more outputs of the second subprocess based on the values of the one or more manipulated variables;” (Devarkonda teaches multiple component models i.e. subprocess models providing predictions in Devarkonda [0041] "Moreover, each of the upstream machine-learning models provides predictions to both a final-assembly machine-learning model 122 and a causal-analysis machine-learning model 124. These predictions could relate to expected production times, expected delay times, expected production amounts, etc. of raw materials, parts, components, subassemblies, process steps, and/or the like. The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150."), and
“and use the main process model to predict future values of the one or more controlled variables based on the future values of the one or more first forecast variables predicted by the first subprocess model and the future values of the one or more second forecast variables predicted by the second subprocess model.” (Devarkonda teaches a final-assembly model i.e. main process model providing product predictions based on the predictions of the component models in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120. " and in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation.").
Claim 8:
Devarkonda teaches “A method for controlling operation of a plant comprising a subprocess and a main process affected by the subprocess,” (Devarkonda teaches a final-assembly model 122 which predicts product-throughput i.e. it is a main process and that it is affected by first, second, and third component models i.e. subprocesses in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120." and Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."),
“the method comprising: obtaining actual historical values of one or more forecast variables representing one or more outputs of the subprocess, actual historical values of one or more controlled variables representing one or more outputs of the main process, and historical values of one or more manipulated variables representing one or more inputs to the plant;” (Devarkonda teaches the embodiments disclosed involve prediction of production capacity using production data from multiple levels of a production stream i.e. historical values of outputs of subprocesses and main processes in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources."; Devarkonda teaches data stores that contain historical data in Devarkonda [0039] "Each of the data stores provides historical data relevant to the component produced by the corresponding factory. This historical data can include data items such as raw materials used, amount of material used i.e. inputs to the plant, manufacturing steps, configuration of the manufacturing steps, actual duration of the manufacturing steps, quality metrics, energy expense, labor expense, production capacity of equipment, time taken to transport raw materials or parts i.e. historical values of outputs of subprocesses and/or any one or more other data items deemed suitable by those of skill in the art for a given implementation.... In at least one embodiment, each of the first-component historical data 140, the second-component historical data 142, and the third-component historical data 144 includes historical data pertaining to the corresponding operational metrics of the corresponding factory. i.e. historical values of outputs of the main process" (emphasis added)),
“training a subprocess model representing the subprocess” (Devarkonda teaches that the models are trained using historical observations of production metrics in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources. Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans. Some embodiments involve inferring, using the machine-learning models and the observed production operational metrics, causal factors that impact the predicted production operational metrics. Some embodiments also involve generation of prognostic alerts, action recommendations based on simulation and generative models, and/or predicted value impacts for a plurality of actions related to the operation of the production value stream. As used herein, “causal factors” include those factors that are elements that contribute in some manner to (e.g., determine) the outcome of a process."),
“training a main process model representing the main process” (Devarkonda teaches a final-assembly model 122 which predicts product-throughput i.e. it represents a main process in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches that the models are trained using historical observations of production metrics and plans derived from the production value stream i.e. historical values of controlled variables in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources. Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans. Some embodiments involve inferring, using the machine-learning models and the observed production operational metrics, causal factors that impact the predicted production operational metrics. Some embodiments also involve generation of prognostic alerts, action recommendations based on simulation and generative models, and/or predicted value impacts for a plurality of actions related to the operation of the production value stream. As used herein, “causal factors” include those factors that are elements that contribute in some manner to (e.g., determine) the outcome of a process."),
“executing a predictive control process using the subprocess model and the main process model to control operation of the plant,” (Devarkonda teaches an action-and-alert process that provides recommended actions 168 to an implementation interface 160 in Devarkonda [0043] "The causal-analysis machine-learning model 124 provides identified causal factor 154 to an action-and-alert process 126. Based at least in part on the identified causal factor 154, the action-and-alert process 126 generates alerts 156 and recommended actions 168, and transmits both to the implementation interface 160, which could include one or more user interfaces for human users (e.g., graphical user interfaces (GUIs), audiovisual interfaces, and/or the like) and/or one or more automated interfaces for carrying out automated processing of the alerts 156 and recommended actions 168."; Devarkonda teaches that the implementation interface transmits commands to alter operating parameters i.e. control operation of the plant in Devarkonda [0046] "The implementation interface 160 generates implementation commands 162 and transmits the implementation commands 162 to the final-assembly/production process 108 in order to alter one or more operating parameters of the final-assembly process 108. In some embodiments, although not pictured in FIG. 1, the implementation interface 160 also or instead generates commands for transmission to one, some, or all of the upstream nodes (e.g., the first-component factory 102, the second-component factory 104, and/or the third-component factory 106), in order to alter one or more operating parameters of the operations there."),
“the predictive control process comprising: using the subprocess model to predict future values of the one or more forecast variables based on values of the one or more manipulated variables;” (Devarkonda teaches a first-component machine learning model i.e. subprocess model which provides first-component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150. Any machine-learning model described herein can be implemented using a machine-learning program executing on one or more hardware platforms, and can be operated as a standalone machine-learning model and/or be combined with one or more other machine-learning models in various different embodiments."; Devarkonda teaches that the models make predictions based on observed production metrics i.e. manipulated variables in Devarkonda [0035] "Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."),
“using the main process model to predict future values of the one or more controlled variables based on the future values of the one or more forecast variables predicted by the subprocess model;” (Devarkonda teaches the final-assembly model receives component predictions from the first component model in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches the final-assembly model generates product-level predictions in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."), and
“and controlling operation of the plant based on the future values of the one or more controlled variables predicted by the main process model.” (Devarkonda teaches a causal-analysis model that receives the predictions from the first-component model and final-assembly model and provides a causal factor to an action-and-alert process which generates recommended actions in Devarkonda [0043] "As described, the causal-analysis machine-learning model 124 also receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120. Based on all of those sets of predictions, the causal-analysis machine-learning model 124 infers causal factor of one or more of the predictions generated by one or more of the first-component machine-learning model 116, the second-component machine-learning model 118, the third-component machine-learning model 120, and the final-assembly machine-learning model 122. The causal-analysis machine-learning model 124 provides identified causal factor 154 to an action-and-alert process 126. Based at least in part on the identified causal factor 154, the action-and-alert process 126 generates alerts 156 and recommended actions 168, and transmits both to the implementation interface 160"; Devarkonda teaches the implementation interface 160 may alter operating parameters of the process in Devarkonda [0046] "The implementation interface 160 generates implementation commands 162 and transmits the implementation commands 162 to the final-assembly/production process 108 in order to alter one or more operating parameters of the final-assembly process 108.”).
Devarkonda does not appear to explicitly teach “train a subprocess model representing the subprocess, wherein training the subprocess model comprises (i) using the subprocess model to generate predicted historical values of the one or more forecast variables as a function of prior predicted historical values of the one or more forecast variables generated by the subprocess model and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more forecast variables generated by the subprocess model with the actual historical values of the one or more forecast variables;” or “train a main process model representing the main process, wherein training the main process model comprises (i) using the main process model to generate predicted historical values of the one or more controlled variables as a function of prior predicted historical values of the one or more controlled variables generated by the main process model, the actual historical values of the one or more forecast variables, and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more controlled variables generated by the main process model with the actual historical values of the one or more controlled variables;” however, Gupta does teach these claim limitation.
Gupta teaches “train a subprocess model representing the subprocess, wherein training the subprocess model comprises (i) using the subprocess model to generate predicted historical values of the one or more forecast variables as a function of prior predicted historical values of the one or more forecast variables generated by the subprocess model and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more forecast variables generated by the subprocess model with the actual historical values of the one or more forecast variables;” (Gupta teaches a model where a predicted output is provided back to the model as input in order to predict the production of a component in a next time step in addition to prior production data i.e. historical values of controlled variables in Gupta [Page 5, first two paragraphs] "The ML-based forecasts were generated from an autoregressive ensemble model. This model accepts as inputs well completion parameters, subsurface geology, inter-well spacing, and historical production, when available. Inputs are subject to standard preprocessing (e.g. encoding any categorical data as one hot vectors), and missing values are imputed, but there are no outlier removal or correction steps. Feature selection is used to choose the n best performing features, where n is determined empirically via tuning. The final regressor used was an extremely randomized trees (extra-trees or ET) model, which was chosen in part for its robust handling of noise. The model output is a per-well timeseries of actual and forecasted production, indexed by IP day (days since well inception). For a given well, the model starts forecasting at the first timepoint t, using the production history from a fixed-size window into the past, t-w:t, and predicting values up to a fixed-size horizon, t:t+h. The model then moves h steps forward in time, and uses its own prior predictions along with older production history t+h-w:t+h to generate a forecast over the next horizon, t+h:t+2h."; Gupta teaches the model is tested by holding back data during training and comparing the predicted values to the actual values in Gupta [Page 6, Evaluation Metrics section, first paragraph] "Standard benchmarks were set for machine learning forecasts which must be met before machine learning can be integrated in traditional forecasting processes. 80% of the total well set was taken for training the models and the remaining 20% was withheld for testing the trained models. A series of chronological back tests, also called time machine testing, were used to evaluate the short-term and long-term performance of the machine learning models. For time machine testing, production data for all the training wells for up to 2 years from the current production date was held back i.e. not exposed to the machine learning models for training. The model predictions were then compared with this held back production to test the model performance."), and
“train a main process model representing the main process, wherein training the main process model comprises (i) using the main process model to generate predicted historical values of the one or more controlled variables as a function of prior predicted historical values of the one or more controlled variables generated by the main process model, the actual historical values of the one or more forecast variables, and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more controlled variables generated by the main process model with the actual historical values of the one or more controlled variables;” (Gupta teaches a model where a predicted output is provided back to the model as input in order to predict the production of a component in a next time step in addition to prior production data i.e. historical values of controlled variables in Gupta [Page 5, first two paragraphs] "The ML-based forecasts were generated from an autoregressive ensemble model. This model accepts as inputs well completion parameters, subsurface geology, inter-well spacing, and historical production, when available. Inputs are subject to standard preprocessing (e.g. encoding any categorical data as one hot vectors), and missing values are imputed, but there are no outlier removal or correction steps. Feature selection is used to choose the n best performing features, where n is determined empirically via tuning. The final regressor used was an extremely randomized trees (extra-trees or ET) model, which was chosen in part for its robust handling of noise. The model output is a per-well timeseries of actual and forecasted production, indexed by IP day (days since well inception). For a given well, the model starts forecasting at the first timepoint t, using the production history from a fixed-size window into the past, t-w:t, and predicting values up to a fixed-size horizon, t:t+h. The model then moves h steps forward in time, and uses its own prior predictions along with older production history t+h-w:t+h to generate a forecast over the next horizon, t+h:t+2h."; Gupta teaches the model is tested by holding back data during training and comparing the predicted values to the actual values in Gupta [Page 6, Evaluation Metrics section, first paragraph] "Standard benchmarks were set for machine learning forecasts which must be met before machine learning can be integrated in traditional forecasting processes. 80% of the total well set was taken for training the models and the remaining 20% was withheld for testing the trained models. A series of chronological back tests, also called time machine testing, were used to evaluate the short-term and long-term performance of the machine learning models. For time machine testing, production data for all the training wells for up to 2 years from the current production date was held back i.e. not exposed to the machine learning models for training. The model predictions were then compared with this held back production to test the model performance."; The concept of training the model using its own predictions as inputs may be applied to the model of Devarkonda which a person having ordinary skill in the art would recognize as an improvement to the model.).
Devarkonda and Gupta are analogous art because they are from the same field of endeavor of making predictions using models. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having teachings of Devarkonda and Gupta before him/her, to modify the teachings of Systems and methods for automating production intelligence across value streams using interconnected machine-learning models of Devarkonda to include the training and validation of a model by using both known inputs and previously predicted outputs as inputs to a model of Gupta because adding the Autoregressive and Machine Learning Driven Production Forecasting of Gupta would allow for improved forecasting as described in Gupta [Pages 11-12, first 2 paragraphs of “Discussion” section] “This successfully demonstrated the potential of ML-based approach as a practical replacement for Arps based forecasting technique. Although the primary objective of the pilot was to test the accuracy of the ML-based techniques, we strongly believe the biggest value will come from workflow automation and efficiency. In a market where efficiencies are crucial for the organization’s bottom-line, our mission is to achieve the best results. In the pre-drill models, the largest relative errors occur during early time for the oil phase, which may result from operating conditions for a well such as tank battery constraints, artificial lift selection, and artificial lift intensity. Over time, pre-drill relative errors stabilized at 2-4% range and became asymptotic. Emphasis was placed on improving oil phase forecasts as the primary driver for economic value during the life of the well; therefore, oil phase models yield better results than the other two phases. In the future, additional work will be performed to elevate the accuracy of gas and water phases. In the post-drill models, satisfactory accuracy levels were also obtained. During two years of production, the program level error stayed within +/- 5%, which was better than traditional Arps for the same well set. This outperformance can be attributed in part to ML-based approach’s flexibility to not follow a defined equation.”
Claims 9-13:
The limitations of claim 9-13 are substantially the same as claims 2-6 respectively and they are rejected for the same reasons.
Claim 15:
Devarkonda teaches “A predictive control system for a plant comprising a subprocess and a main process affected by the subprocess,” (Devarkonda teaches a final-assembly model 122 which predicts product-throughput i.e. it is a main process and that it is affected by first, second, and third component models i.e. subprocesses in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120." and Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."),
“the predictive control system comprising one or more processing circuits configured to: independently train a subprocess model representing the subprocess and a main process model representing the main process without using predictions generated by the subprocess model when training the main process model;” (Devarkonda teaches a machine which instructions that cause it to perform the methods discussed in Devarkonda [0133] "FIG. 16 is a diagrammatic representation of a machine 1600 within which instructions 1612 (e.g., software, a program, an application, an applet, an app, or other executable code) for causing the machine 1600 to perform any one or more of the methodologies discussed herein may be executed." and in Devarkonda [0134] "The machine 1600 may include processors 1602, memory 1604, and I/O components 1606, which may be configured to communicate with each other via a bus 1644."; Devarkonda teaches that the models may be trained using supervised learning and ran against a training dataset in Devarkonda [0108-0109] "During a learning phase, the models are developed against a training dataset of inputs to optimize the models to correctly predict the output for a given input. Generally, the learning phase may be supervised, semi-supervised, or unsupervised; indicating a decreasing level to which the “correct” outputs are provided in correspondence to the training inputs. In a supervised learning phase, all of the outputs are provided to the model and the model is directed to develop a general rule or algorithm that maps the input to the output... Models may be run against a training dataset for several epochs (e.g., iterations), in which the training dataset is repeatedly fed into the model to refine its results. For example, in a supervised learning phase, a model is developed to predict the output for a given set of inputs, and is evaluated over several epochs to more reliably provide the output that is specified as corresponding to the given input for the greatest number of inputs for the training dataset."; Devarkonda teaches that the models are trained using historical observations of production metrics i.e. it doesn't use predictions generated by the subprocess model in Devarkonda [0035] "The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."),
“after independently training the subprocess model and the main process model, connect the subprocess model to the main process model such that the predictions generated by the subprocess model are provided as inputs to the main process model;” (Devarkonda teaches the final-assembly model receives component predictions from the first component model in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches the final-assembly model generates product-level predictions in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."),
“execute a predictive control process using the subprocess model to generate predicted future values of the one or more forecast variables” (Devarkonda teaches a first-component machine learning model i.e. subprocess model which provides first-component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150. Any machine-learning model described herein can be implemented using a machine-learning program executing on one or more hardware platforms, and can be operated as a standalone machine-learning model and/or be combined with one or more other machine-learning models in various different embodiments."; Devarkonda teaches that the models make predictions based on observed production metrics in Devarkonda [0035] "Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."),
“and using the main process model to generate predicted future values of the one or more controlled variables based on the predicted future values of the one or more forecast variables generated by the subprocess model;” (Devarkonda teaches the final-assembly model receives component predictions from the first component model in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches the final-assembly model generates product-level predictions in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."), and
“and control operation of the plant using the predicted values of the one or more controlled variables generated by the main process model.” (Devarkonda teaches a causal-analysis model that receives the predictions from the first-component model and final-assembly model and provides a causal factor to an action-and-alert process which generates recommended actions in Devarkonda [0043] "As described, the causal-analysis machine-learning model 124 also receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120. Based on all of those sets of predictions, the causal-analysis machine-learning model 124 infers causal factor of one or more of the predictions generated by one or more of the first-component machine-learning model 116, the second-component machine-learning model 118, the third-component machine-learning model 120, and the final-assembly machine-learning model 122. The causal-analysis machine-learning model 124 provides identified causal factor 154 to an action-and-alert process 126. Based at least in part on the identified causal factor 154, the action-and-alert process 126 generates alerts 156 and recommended actions 168, and transmits both to the implementation interface 160"; Devarkonda teaches the implementation interface 160 may alter operating parameters of the process in Devarkonda [0046] "The implementation interface 160 generates implementation commands 162 and transmits the implementation commands 162 to the final-assembly/production process 108 in order to alter one or more operating parameters of the final-assembly process 108. ").
Devarkonda does not appear to explicitly teach “wherein: training the subprocess model comprises (i) using the subprocess model to generate predicted historical values of one or more forecast variables as a function of prior predicted historical values of the one or more forecast variables generated by the subprocess model and historical values of one or more manipulated variables and (ii) comparing the predicted historical values of the one or more forecast variables generated by the subprocess model with actual historical values of the one or more forecast variables;” or “and training the main process model comprises (i) using the main process model to generate predicted historical values of one or more controlled variables as a function of prior predicted historical values of the one or more controlled variables generated by the main process model, the actual historical values of the one or more forecast variables, and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more controlled variables generated by the main process model with actual historical values of the one or more controlled variables;” however, Gupta does teach these claim limitation.
Gupta teaches “wherein: training the subprocess model comprises (i) using the subprocess model to generate predicted historical values of one or more forecast variables as a function of prior predicted historical values of the one or more forecast variables generated by the subprocess model and historical values of one or more manipulated variables and (ii) comparing the predicted historical values of the one or more forecast variables generated by the subprocess model with actual historical values of the one or more forecast variables;” (Gupta teaches a model where a predicted output is provided back to the model as input in order to predict the production of a component in a next time step in addition to prior production data i.e. historical values of controlled variables in Gupta [Page 5, first two paragraphs] "The ML-based forecasts were generated from an autoregressive ensemble model. This model accepts as inputs well completion parameters, subsurface geology, inter-well spacing, and historical production, when available. Inputs are subject to standard preprocessing (e.g. encoding any categorical data as one hot vectors), and missing values are imputed, but there are no outlier removal or correction steps. Feature selection is used to choose the n best performing features, where n is determined empirically via tuning. The final regressor used was an extremely randomized trees (extra-trees or ET) model, which was chosen in part for its robust handling of noise. The model output is a per-well timeseries of actual and forecasted production, indexed by IP day (days since well inception). For a given well, the model starts forecasting at the first timepoint t, using the production history from a fixed-size window into the past, t-w:t, and predicting values up to a fixed-size horizon, t:t+h. The model then moves h steps forward in time, and uses its own prior predictions along with older production history t+h-w:t+h to generate a forecast over the next horizon, t+h:t+2h."; Gupta teaches the model is tested by holding back data during training and comparing the predicted values to the actual values in Gupta [Page 6, Evaluation Metrics section, first paragraph] "Standard benchmarks were set for machine learning forecasts which must be met before machine learning can be integrated in traditional forecasting processes. 80% of the total well set was taken for training the models and the remaining 20% was withheld for testing the trained models. A series of chronological back tests, also called time machine testing, were used to evaluate the short-term and long-term performance of the machine learning models. For time machine testing, production data for all the training wells for up to 2 years from the current production date was held back i.e. not exposed to the machine learning models for training. The model predictions were then compared with this held back production to test the model performance."), and
“and training the main process model comprises (i) using the main process model to generate predicted historical values of one or more controlled variables as a function of prior predicted historical values of the one or more controlled variables generated by the main process model, the actual historical values of the one or more forecast variables, and the historical values of the one or more manipulated variables and (ii) comparing the predicted historical values of the one or more controlled variables generated by the main process model with actual historical values of the one or more controlled variables;” (Gupta teaches a model where a predicted output is provided back to the model as input in order to predict the production of a component in a next time step in addition to prior production data i.e. historical values of controlled variables in Gupta [Page 5, first two paragraphs] "The ML-based forecasts were generated from an autoregressive ensemble model. This model accepts as inputs well completion parameters, subsurface geology, inter-well spacing, and historical production, when available. Inputs are subject to standard preprocessing (e.g. encoding any categorical data as one hot vectors), and missing values are imputed, but there are no outlier removal or correction steps. Feature selection is used to choose the n best performing features, where n is determined empirically via tuning. The final regressor used was an extremely randomized trees (extra-trees or ET) model, which was chosen in part for its robust handling of noise. The model output is a per-well timeseries of actual and forecasted production, indexed by IP day (days since well inception). For a given well, the model starts forecasting at the first timepoint t, using the production history from a fixed-size window into the past, t-w:t, and predicting values up to a fixed-size horizon, t:t+h. The model then moves h steps forward in time, and uses its own prior predictions along with older production history t+h-w:t+h to generate a forecast over the next horizon, t+h:t+2h."; Gupta teaches the model is tested by holding back data during training and comparing the predicted values to the actual values in Gupta [Page 6, Evaluation Metrics section, first paragraph] "Standard benchmarks were set for machine learning forecasts which must be met before machine learning can be integrated in traditional forecasting processes. 80% of the total well set was taken for training the models and the remaining 20% was withheld for testing the trained models. A series of chronological back tests, also called time machine testing, were used to evaluate the short-term and long-term performance of the machine learning models. For time machine testing, production data for all the training wells for up to 2 years from the current production date was held back i.e. not exposed to the machine learning models for training. The model predictions were then compared with this held back production to test the model performance."; The concept of training the model using its own predictions as inputs may be applied to the model of Devarkonda which a person having ordinary skill in the art would recognize as an improvement to the model.).
Devarkonda and Gupta are analogous art because they are from the same field of endeavor of making predictions using models. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having teachings of Devarkonda and Gupta before him/her, to modify the teachings of Systems and methods for automating production intelligence across value streams using interconnected machine-learning models of Devarkonda to include the training and validation of a model by using both known inputs and previously predicted outputs as inputs to a model of Gupta because adding the Autoregressive and Machine Learning Driven Production Forecasting of Gupta would allow for improved forecasting as described in Gupta [Pages 11-12, first 2 paragraphs of “Discussion” section] “This successfully demonstrated the potential of ML-based approach as a practical replacement for Arps based forecasting technique. Although the primary objective of the pilot was to test the accuracy of the ML-based techniques, we strongly believe the biggest value will come from workflow automation and efficiency. In a market where efficiencies are crucial for the organization’s bottom-line, our mission is to achieve the best results. In the pre-drill models, the largest relative errors occur during early time for the oil phase, which may result from operating conditions for a well such as tank battery constraints, artificial lift selection, and artificial lift intensity. Over time, pre-drill relative errors stabilized at 2-4% range and became asymptotic. Emphasis was placed on improving oil phase forecasts as the primary driver for economic value during the life of the well; therefore, oil phase models yield better results than the other two phases. In the future, additional work will be performed to elevate the accuracy of gas and water phases. In the post-drill models, satisfactory accuracy levels were also obtained. During two years of production, the program level error stayed within +/- 5%, which was better than traditional Arps for the same well set. This outperformance can be attributed in part to ML-based approach’s flexibility to not follow a defined equation.”
Claim 16:
Devarkonda in view of Gupta teaches “The predictive control system of claim 15, wherein training the subprocess model comprises: using the historical values of the one or more manipulated variables to generate the predicted historical values of the one or more forecast variables using the subprocess model;” (Devarkonda teaches the embodiments disclosed involve prediction of production capacity using production data from multiple levels of a production stream i.e. historical values of outputs of subprocesses and main processes in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources."; Devarkonda teaches data stores that contain historical data in Devarkonda [0039] "Each of the data stores provides historical data relevant to the component produced by the corresponding factory. This historical data can include data items such as raw materials used, amount of material used i.e. inputs to the plant, manufacturing steps, configuration of the manufacturing steps, actual duration of the manufacturing steps, quality metrics, energy expense, labor expense, production capacity of equipment, time taken to transport raw materials or parts i.e. historical values of outputs of subprocesses and/or any one or more other data items deemed suitable by those of skill in the art for a given implementation.... In at least one embodiment, each of the first-component historical data 140, the second-component historical data 142, and the third-component historical data 144 includes historical data pertaining to the corresponding operational metrics of the corresponding factory. i.e. historical values of outputs of the main process" (emphasis added); Devarkonda teaches that machine learning algorithms use training data to train the program in Devarkonda [0105-0106] "The machine-learning algorithms utilize the training data 1312 to find correlations among the identified features 1302 that affect the outcome or assessments 1318. In some example embodiments, the training data 1312 includes labeled data, which is known data for one or more identified features 1302 and one or more outcomes, such as predicted upstream and/or end-production delays, predicted throughput of a production value stream, and/or the like. With the training data 1312 and the identified features 1302, the machine-learning tool is trained at machine-learning-program training operation 1310. The machine-learning tool appraises the value of the features 1302 as they correlate to the training data 1312. The result of the training is the trained machine-learning program 1314."), and
“and adjusting the subprocess model to reduce an error between the predicted historical values of the one or more forecast variables and the actual historical values of the one or more forecast variables.” (Devarkonda teaches that backpropagation may be used in order to reduce an error value in Devarkonda [0118-0119] "In training of a DNN architecture, a regression, which is structured as a set of statistical processes for estimating the relationships among variables, can include a minimization of a cost function. The cost function may be implemented as a function to return a number representing how well the neural network performed in mapping training examples to correct output. In training, if the cost function value is not within a pre-determined range, based on the known training images, backpropagation is used, where backpropagation is a common method of training artificial neural networks that are used with an optimization method such as a stochastic gradient descent (SGD) method. Use of backpropagation can include propagation and weight update. When an input is presented to the neural network, it is propagated forward through the neural network, layer by layer, until it reaches the output layer. The output of the neural network is then compared to the desired output, using the cost function, and an error value is calculated for each of the nodes in the output layer. The error values are propagated backwards, starting from the output, until each node has an associated error value which roughly represents its contribution to the original output. Backpropagation can use these error values to calculate the gradient of the cost function with respect to the weights in the neural network.").
Claim 17:
Devarkonda in view of Gupta teaches “The predictive control system of claim 15, wherein training the main process model comprises: using the actual historical values of the one or more forecast variables and the historical values of one or more manipulated variables to generate the predicted historical values of the one or more controlled variables using the main process model;” (Devarkonda teaches a final-assembly model 122 which predicts product-throughput i.e. it represents a main process in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120."; Devarkonda teaches that the models are trained using historical observations of production metrics and plans derived from the production value stream i.e. historical values of forecast and manipulated variables in Devarkonda [0035] "One or more disclosed embodiments involve prediction of production capacity for industrial manufacturing operations (both continuous and discrete) based on disparate data sources. Some embodiments involve receiving production data for multiple levels of a production value stream. This production data can include input signals related to a variety of production components, located across the production value stream. The input signals may relate to observed production operational metrics and plans derived from the production value stream. As examples, production planning and process data may be received for the production of raw materials, parts, components, sub-assemblies, and/or modules at specific plants and tagged to specific production lines and resources. Based on the observed production operational metrics and planned production certain operational metrics are predicted. The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans. Some embodiments involve inferring, using the machine-learning models and the observed production operational metrics, causal factors that impact the predicted production operational metrics. Some embodiments also involve generation of prognostic alerts, action recommendations based on simulation and generative models, and/or predicted value impacts for a plurality of actions related to the operation of the production value stream. As used herein, “causal factors” include those factors that are elements that contribute in some manner to (e.g., determine) the outcome of a process."), and
“and adjusting the main process model to reduce an error between the predicted historical values of the one or more controlled variables and the actual historical values of the one or more controlled variables.” (Devarkonda teaches that backpropagation may be used in order to reduce an error value in Devarkonda [0118-0119] "In training of a DNN architecture, a regression, which is structured as a set of statistical processes for estimating the relationships among variables, can include a minimization of a cost function. The cost function may be implemented as a function to return a number representing how well the neural network performed in mapping training examples to correct output. In training, if the cost function value is not within a pre-determined range, based on the known training images, backpropagation is used, where backpropagation is a common method of training artificial neural networks that are used with an optimization method such as a stochastic gradient descent (SGD) method. Use of backpropagation can include propagation and weight update. When an input is presented to the neural network, it is propagated forward through the neural network, layer by layer, until it reaches the output layer. The output of the neural network is then compared to the desired output, using the cost function, and an error value is calculated for each of the nodes in the output layer. The error values are propagated backwards, starting from the output, until each node has an associated error value which roughly represents its contribution to the original output. Backpropagation can use these error values to calculate the gradient of the cost function with respect to the weights in the neural network.").
Claim 18:
The limitations of claim 18 are substantially the same as claim 3 and it is rejected for the same reasons.
Claim 19:
Devarkonda in view of Gupta teaches “The predictive control system of claim 15, wherein: training the subprocess model and the main process model comprises independently training a plurality of subprocess models representing a plurality of subprocesses of the plant and a plurality of main process models representing a plurality of main processes of the plant;” (Devarkonda teaches that the models may be trained using supervised learning and ran against a training dataset in Devarkonda [0108-0109] "During a learning phase, the models are developed against a training dataset of inputs to optimize the models to correctly predict the output for a given input. Generally, the learning phase may be supervised, semi-supervised, or unsupervised; indicating a decreasing level to which the “correct” outputs are provided in correspondence to the training inputs. In a supervised learning phase, all of the outputs are provided to the model and the model is directed to develop a general rule or algorithm that maps the input to the output... Models may be run against a training dataset for several epochs (e.g., iterations), in which the training dataset is repeatedly fed into the model to refine its results. For example, in a supervised learning phase, a model is developed to predict the output for a given set of inputs, and is evaluated over several epochs to more reliably provide the output that is specified as corresponding to the given input for the greatest number of inputs for the training dataset."; Devarkonda teaches that the models are trained using historical observations of production metrics i.e. it doesn't use predictions generated by the subprocess model in Devarkonda [0035] "The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."), and
“and connecting the subprocess model to the main process model comprises connecting a first subprocess model of the plurality of subprocess models and a second subprocess model of the plurality of subprocess models in series with each other and with one or more of the plurality of main process models such that: a first output of the first subprocess model is provided as an input to the second subprocess model; and a second output of the second subprocess model is provided as an input to the one or more of the plurality of main process models.” (Devarkonda teaches component models providing component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150"; Devarkonda teaches a final-assembly model providing product predictions i.e. the final-assembly model may be regarded as a subassembly model which provides an input to the causal-analysis model in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."; Devarkonda teaches that models may be deployed in connection with raw-material providers, sub-assembly providers, and others i.e. there may be models providing information to the component models in Devarkonda [0040] "The three example upstream machine-learning models that are depicted in FIG. 1 are examples of upstream machine-learning models that correspond to providers of components that are later combined into an end product. This is one example of a type of upstream machine-learning model that can be deployed in embodiments of the present disclosure. Other examples include machine-learning models deployed in connection with raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like. Other examples include machine-learning models deployed in connection with process manufacturing raw-material providers, processed-material providers, sub-assembly (e.g., module) providers that, e.g., combine parts into sub-assemblies that are ultimately further combined into an end product, process steps, and/or the like.").
Claim 20:
Devarkonda in view of Gupta teaches “The predictive control system of claim 15, wherein: training the subprocess model and the main process model comprises independently training a plurality of subprocess models representing a plurality of subprocesses of the plant and a plurality of main process models representing a plurality of main processes of the plant;” (Devarkonda teaches that the models may be trained using supervised learning and ran against a training dataset in Devarkonda [0108-0109] "During a learning phase, the models are developed against a training dataset of inputs to optimize the models to correctly predict the output for a given input. Generally, the learning phase may be supervised, semi-supervised, or unsupervised; indicating a decreasing level to which the “correct” outputs are provided in correspondence to the training inputs. In a supervised learning phase, all of the outputs are provided to the model and the model is directed to develop a general rule or algorithm that maps the input to the output... Models may be run against a training dataset for several epochs (e.g., iterations), in which the training dataset is repeatedly fed into the model to refine its results. For example, in a supervised learning phase, a model is developed to predict the output for a given set of inputs, and is evaluated over several epochs to more reliably provide the output that is specified as corresponding to the given input for the greatest number of inputs for the training dataset."; Devarkonda teaches that the models are trained using historical observations of production metrics i.e. it doesn't use predictions generated by the subprocess model in Devarkonda [0035] "The predictions may be performed by interconnected machine-learning models. In at least some embodiments, the machine-learning models are trained based on historical observations of production operational metrics and latest production scheduling and plans."),
“and connecting the subprocess model to the main process model comprises connecting a first subprocess model of the plurality of subprocess models and a second subprocess model of the plurality of subprocess models in parallel with each other and in series with one or more of the plurality of main process models” (Devarkonda teaches component models providing component predictions in Devarkonda [0041] "The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150"; Devarkonda teaches a final-assembly model providing product predictions based on the predictions of the component models in Devarkonda [0042] "The final-assembly machine-learning model 122 receives the first-component predictions 146 from the first-component machine-learning model 116, the second-component predictions 148 from the second-component machine-learning model 118, and the third-component predictions 150 from the third-component machine-learning model 120. " and in Devarkonda [0043] "The final-assembly machine-learning model 122 generates and transmits product-level predictions 152 to the causal-analysis machine-learning model 124. The product-level predictions 152 could relate to expected production levels, expected production amounts, expected production times, expected production delays, expected production costs, and/or any one or more other types of product-level predictions 152 deemed suitable by those of skill in the art for a given implementation."), and
“such that: a first output of the first subprocess model is provided as a first input to the one or more of the plurality of main process models; and a second output of the second subprocess model is provided as a second input to the one or more of the plurality of main process models.” (Devarkonda teaches multiple component models i.e. subprocess models providing predictions in Devarkonda [0041] "Moreover, each of the upstream machine-learning models provides predictions to both a final-assembly machine-learning model 122 and a causal-analysis machine-learning model 124. These predictions could relate to expected production times, expected delay times, expected production amounts, etc. of raw materials, parts, components, subassemblies, process steps, and/or the like. The first-component machine-learning model 116 provides first-component predictions 146, the second-component machine-learning model 118 provides second-component predictions 148, and the third-component machine-learning model 120 provides third-component predictions 150.").
Claims 7 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Devarkonda (US20220067622A1), in view of Gupta, further in view of Discenzo (US20140156050A1).
Claim 7:
Devarkonda in view of Gupta teaches “The predictive control system of claim 1,” as described above. Devarkonda and Gupta do not appear to explicitly teach “wherein controlling operation of the plant comprises: using the values of the one or more controlled variables predicted by the main process model to train a controller neural network model during an offline training phase of the predictive control process;” or “and using the controller neural network model to generate values of the one or more manipulated variables during an online operation phase of the predictive control process.” However, Discenzo does teach these claim limitations.
Discenzo teaches “wherein controlling operation of the plant comprises: using the values of the one or more controlled variables predicted by the main process model to train a controller neural network model during an offline training phase of the predictive control process;” (Discenzo teaches that a neural network estimator may be used that was trained with off-line data in Discenzo [0038] "In a further example, a neural network estimator for real-time bio-assessment is utilized (e.g., employ an autoassociative neural network that has been trained using at-line or off-line laboratory data). In accordance with some aspects, a combination of these techniques can be utilized. In accordance with some aspects, the use of autoassociative neural networks can be utilized to confirm proper sensor readings and/or to estimate other readings on-line. In accordance with some aspects, the autoassociative neural network may synthesize the value of what would otherwise be returned from a defective sensor. The autoassociative neural networks can execute in real-time on a controller and/or can be trained off-line using laboratory analysis results and/or stored sensor data."; Discenzo teaches neural net controllers that may be in a hierarchy in Discenzo [0064] "The controller may be a linear controller, adaptive controller, model-based controller, predictive controller, state-space controller, neural net controller, or other controller topology and may include process (or plant) models and may dynamically change the process based on the sensed or predicted process state. The controllers can be a hierarchy of controllers of a network of controllers, as shown by the exemplary organization information and manufacturing system hierarchy 600 of FIG. 6. In the example hierarchy 600, an enterprise computer system 602 controls a plant controller/scheduler 604, which controls a multitude of area controllers 606 (e.g., PLC), which can control one or more machines 608 and/or sensors 610."), and
“and using the controller neural network model to generate values of the one or more manipulated variables during an online operation phase of the predictive control process.” (Discenzo teaches that a predicted value can be used to change processing i.e. the model is online and generates manipulated variables in Discenzo [0053] "In accordance with some aspects, the autonomous agent corresponds to a process stage, process step, or a group of processes. The at least one autonomous agent predicts the bacteria state of material leaving the process stage. The predicted value can be used to change the processing before the stage is complete and/or the predicted value at the end of the stage is communicated to other upstream and/or downstream process agents.").
Devarkonda, Gupta, and Discenzo are analogous art because they are from the same field of endeavor of predicting production processes using artificial intelligence. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having teachings of Devarkonda, Gupta, and Discenzo before him/her, to modify the teachings of Systems and methods for automating production intelligence across value streams using interconnected machine-learning models of Devarkonda modified to include the training and validation of a model by using both known inputs and previously predicted outputs as inputs to a model of Gupta to include the offline training and online generation of manipulated variables of Discenzo because adding the Microbial monitoring and prediction of Discenzo would allow for confirmation of proper sensor readings or to estimate other readings as described in Discenzo [0038] “In a further example, a neural network estimator for real-time bio-assessment is utilized (e.g., employ an autoassociative neural network that has been trained using at-line or off-line laboratory data). In accordance with some aspects, a combination of these techniques can be utilized. In accordance with some aspects, the use of autoassociative neural networks can be utilized to confirm proper sensor readings and/or to estimate other readings on-line. In accordance with some aspects, the autoassociative neural network may synthesize the value of what would otherwise be returned from a defective sensor. The autoassociative neural networks can execute in real-time on a controller and/or can be trained off-line using laboratory analysis results and/or stored sensor data.”
Claim 14:
The limitations of claim 14 are substantially the same as claim 7 and it is rejected for the same reasons.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
John et al. (US20220266166A1) teaches a method of training a neural network model and autoregressive model using input and output parameters in John [0071], with details on the training method in John [0085-0089].
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Zachary A Cain whose telephone number is (571)272-4503. The examiner can normally be reached Mon-Fri 7:00-3:30 CST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kenneth M Lo can be reached at (571) 272-9774. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Z.A.C./ Examiner, Art Unit 2116 /KENNETH M LO/Supervisory Patent Examiner, Art Unit 2116
Prosecution Insights