DETAILED ACTION
The instant application having application number 18/112877 filled on 2/22/23 has a total of 21 claims pending for examination. There are 3 independent claims and 18 dependent claims, all of which are examined below.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statements (IDS) submitted from February 22nd, 2023 through January 29th, 2025 have been considered by the examiner.
Drawings
The drawings were received on 02/22/23. These drawings are accepted.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 2, 13, 18 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 2, 13, and 18 recite “the constrained function optimization.” However, claims 2, 13, and 18 depend from independent claims 1, 12, and 17, respectively, and independent claims 1, 12, and 17 recite “constrained objective function optimization,” not “constrained function optimization.” Accordingly, there is insufficient antecedent basis for the term “the constrained function optimization” in claims 2, 13, and 18, because it is unclear whether the claimed “constrained function optimization” refers to the previously recited “constrained objective function optimization” or to a different optimization. Correction is required.
For examination purposes, the term “constrained function optimization” in claims 2, 13, and 18 will be interpreted as “constrained objective function optimization.”
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-21 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1, Statutory Category:
Yes: Claims 1-11 are directed to a method.
Step 2A Prong I, judicial Exception:
The Examiner submits that the foregoing claim limitations constitute mental processes and mathematical concepts when given their broadest reasonable interpretation. Abstract ideas are bolded.
Claim 1 recites the limitations:
A method for optimizing settings in an industrial process, the method comprising:
providing, on a computer system, a physics-based digital model of an industrial process, the digital model having an input setting associated with a piece of equipment in the industrial process, the digital model configured to generate dependent values based on the input setting, the dependent values including a constrained value and a target value;
providing, on the computer system, a machine learning (ML) model of the industrial process, the ML model having been trained to emulate the physics-based digital model, including the input setting and the dependent values;
generating initial guesses for the input setting;
selecting an initial guess from the initial guesses, the selected initial guess associated with a computational refinement branch, the branch comprising:
maximizing or minimizing the target value produced by the ML model by running the initial guess through the ML model and performing a constrained objective function optimization using the ML model to find a refined guess of the input setting;
inputting the refined guess into the physics-based digital model to check the constrained value;
determining, using the physics-based digital model, that the constrained value violates a constraint;
subjecting the refined guess and violated constrained value to a constraint correction projection to generate a refined input;
inputting the refined input into the physics-based digital model to re-check the constrained value;
verifying that the constrained value no longer violates the constraint based on the re-check; and
logging the refined input as a candidate input setting for the industrial process;
selecting among multiple candidate input settings for a best input setting, each candidate input setting originating from a different initial guess of the initial guesses and a respective computational refinement branch.
The limitation generate dependent values based on the input setting, the dependent values including a constrained value and a target value recites a mental process under MPEP § 2106.04(a)(2)(III) because it can be practically performed in the human mind or with pen and paper through observation, evaluation, and judgment. Specifically, this limitation involves observing an input setting value and making judgments based on known relationships to determine resulting values such as a constrained and target value. For example, a person could look at a temperature input setting associated with a boiling tank in a food production plant and decide what temperature would yield the desired production output, while also judging whether certain temperature values would be outside the safe or permitted bounds of the tank.
The limitation selecting an initial guess from the initial guesses, the selected initial guess associated with a computational refinement branch recites a mental process under MPEP § 2106.04(a)(2)(III) because it can be practically performed in the human mind or with pen and paper through observation, evaluation, and judgment. Specifically, a person could think of multiple possible input setting values, choose one of those values to analyze first, and then use that selected value as the starting point for the rest of the analysis. For example, a person could consider several possible temperature settings for a boiling tank in a food production plant, select one temperature setting as an initial guess, and then proceed with a separate line of analysis for that selected temperature.
The limitation maximizing or minimizing the target value produced by the ML model by running the initial guess through the ML model and performing a constrained objective function optimization using the ML model to find a refined guess of the input setting recites an abstract idea because it is directed to mathematical concepts under MPEP § 2106.04(a)(2)(I), which includes mathematical relationships and mathematical calculations. The claim does not recite how the ML model itself is improved, how the optimization is performed, or how the industrial equipment is actually controlled. Instead, the claim merely uses the ML model like a mathematical function that takes in an initial numerical guess, produces a numerical target value, and is then used to solve a constrained optimization problem to calculate a better numerical guess. This interpretation is consistent with the specification, which describes the objective function as reflecting a relationship between input parameters and dependent parameters and describes the optimization as using mathematical optimization algorithms until convergence or production of an optimal value (e.g. paragraph [0032]).
The limitations inputting the refined guess into the physics-based digital model to check the constrained value and determining, using the physics-based digital model, that the constrained value violates a constraint recite mental processes under MPEP § 2106.04(a)(2)(III) because they can be practically performed in the human mind or with pen and paper through observation, evaluation, and judgment. The claim is recited at a high level of generality and merely requires checking a value and determining whether that value violates a constraint. Because the claim does not recite any particular technological improvement or specific technique for checking the value or determining the constraint violation, the physics-based digital model is only being used as a generic computer-implemented tool to obtain or check the value. Under the broadest reasonable interpretation, a person could read a temperature value for a boiling tank, and compare that value to a known safe operating limit. If the resulting value is above the safe limit, the person could judge that the constrained value violates the constraint.
The limitation subjecting the refined guess and violated constrained value to a constraint correction projection to generate a refined input recites a mental process under MPEP § 2106.04(a)(2)(III) because it can be practically performed in the human mind or with pen and paper through observation, evaluation, and judgment. The limitation merely requires using the refined guess and the violated constrained value to determine a corrected input value. For example, if a person determines that a temperature setting for a boiling tank in a food production plant violates a safe operating limit, the person could make a judgment call to move the temperature setting toward a safer range and use that adjusted temperature as the refined input.
The limitations inputting the refined input into the physics-based digital model to re-check the constrained value and verifying that the constrained value no longer violates the constraint based on the re-check recite mental processes under MPEP § 2106.04(a)(2)(III) because they can be practically performed in the human mind or with pen and paper through observation, evaluation, and judgment. Because the claim does not recite any particular technological improvement or specific way of performing the re-check or verification, the physics-based digital model is only being used as a generic computer-implemented tool to obtain or check the value again. Therefore, the limitation merely involves re-checking a value and determining whether the corrected value is now within the permitted bound. An example of this being performed in the human mind would be a person moving a boiling tank temperature setting toward a safer range, using judgment to determine the resulting temperature or pressure value, and comparing it to the known safe operating limit for the tank. If the resulting value is no longer above the safe limit, the person could judge that the constrained value no longer violates the constraint.
The limitations logging the refined input as a candidate input setting for the industrial process and selecting among multiple candidate input settings for a best input setting, each candidate input setting originating from a different initial guess of the initial guesses and a respective computational refinement branch recite mental processes under MPEP § 2106.04(a)(2)(III) because they can be practically performed in the human mind or with pen and paper through observation, evaluation, and judgment. An example of this would be a person writing down temperature settings for a boiling tank that the person determines are acceptable settings, and then evaluating those written settings to decide which setting best meets the desired production goal.
Step 2A Prong II, Integration into a Practical Application:
Claim 1 recites the following additional claim limitations outside the abstract idea which only present general field or use, mere instructions to apply an exception, and/or insignificant extra solution activity:
A method for optimizing settings in an industrial process, the method comprising (general field of use, see MPEP § 2106.05(h))
providing, on a computer system, a physics-based digital model of an industrial process (mere instructions to apply an exception, see MPEP § 2106.05(f))
the digital model having an input setting associated with a piece of equipment in the industrial process (general field of use, see MPEP § 2106.05(h))
providing, on the computer system, a machine learning (ML) model of the industrial process, the ML model having been trained to emulate the physics-based digital model, including the input setting and the dependent values (mere instructions to apply an exception, see MPEP § 2106.05(f))
inputting the refined guess into the physics-based digital model to check the constrained value (data gathering and organization under MPEP § 2106.05(g) and mere instructions to apply an exception, see MPEP § 2106.05(f))
determining, using the physics-based digital model, that the constrained value violates a constraint (mere instructions to apply an exception, see MPEP § 2106.05(f))
inputting the refined input into the physics-based digital model to re-check the constrained value (data gathering and organization under MPEP § 2106.05(g) and mere instructions to apply an exception, see MPEP § 2106.05(f))
Step 2B, Significantly More:
When considered individually or in combination, the additional limitations and elements of claim 1 do not amount to significantly more than the judicial exceptions. The claim is recited at a high level of generality and merely uses a generic computer system, a physics-based digital model, and an ML model as tools to perform the abstract mathematical optimization and mental evaluation steps. The claim does not recite any particular improvement to the computer system, the ML model, the physics-based digital model, or the industrial equipment. Therefore, the additional limitations amount to no more than applying the abstract idea on generic computer-implemented components in the field of industrial process optimization, and do not provide significantly more than the judicial exceptions.
Regarding claim 2, the claim recites the method of claim 1 wherein the constrained function optimization uses a gradient-based algorithm or a trust region method. Claim 2 does not remove or change the abstract limitations identified in claim 1. Instead, claim 2 merely adds further details to the constrained objective function optimization by specifying that the optimization uses a mathematical technique, such as a gradient-based algorithm or a trust region method. Therefore, claim 2 only provides more specific recitations of the abstract ideas identified in claim 1 and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 2 is not patent eligible.
Regarding claim 3, the claim recites the method of claim 1 wherein the constraint correction projection uses an input parameter perturbation technique. Claim 3 does not remove or change the abstract limitations identified in claim 1. Instead, claim 3 merely adds further details to the constraint correction projection by specifying that the correction uses an input parameter perturbation technique. This further limits the abstract idea by adding a mathematical or statistical operation for adjusting the input value used in the correction step. Therefore, claim 3 only provides more specific recitations of the abstract ideas identified in claim 1 and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 3 is not patent eligible.
Regarding claim 4, the claim recites the method of claim 1 wherein the constraint correction projection includes: determining an amount of violation of the constraint; determining a targeted constraint output for the ML model; computing a gradient of the ML model at the refined guess for the constrained value and the target value; determining a vector direction of the gradient so as to minimally affect the target value; computing a direction of correction for the constrained value; and updating the refined guess based on the computed direction of correction. Claim 4 does not remove or change the abstract limitations identified in claim 1. Instead, claim 4 merely adds further mathematical details to the constraint correction projection by reciting additional steps such as calculating a violation amount, determining a targeted constraint output, computing gradients, determining a vector direction, computing a direction of correction, and updating the refined guess based on that correction. These limitations are directed to mathematical concepts under MPEP § 2106.04(a)(2)(I) because they involve mathematical calculations and relationships, including gradients, vector directions, and correction directions. Therefore, claim 4 only provides more specific recitations of the abstract mathematical optimization and correction steps identified in claim 1, and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 4 is not patent eligible.
Regarding claim 5, the claim recites the method of claim 4 wherein the constraint correction projection is repeated in a loop. Claim 5 does not remove or change the abstract limitations identified in claim 1 or claim 4. Claim 5 only requires repeating the constraint correction projection, which repeats the same mathematical correction steps identified in claim 4, such as calculating a violation amount, computing gradients, determining a correction direction, and updating the refined guess. Repeating the same abstract mathematical calculations does not provide a particular technological improvement or meaningfully limit the judicial exception (see Performing repetitive calculations in MPEP § 2106.05(d)). Therefore, claim 5 does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 5 is not patent eligible.
Regarding claim 6, the claim recites the method of claim 1 further comprising: assessing that a number of times that the physics-based digital model has run is fewer than a limit; and allowing the inputting of the refined input into the physics-based digital model based on the assessing. Claim 6 does not remove or change the abstract limitations identified in claim 1. Instead, claim 6 further recites a mental process because it merely requires counting how many times an action has occurred, comparing that count to a limit, and deciding whether to continue based on the comparison. Therefore, claim 6 only provides more specific recitations of the abstract evaluations and judgments identified in claim 1, and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 6 is not patent eligible.
Regarding claim 7, the claim recites the method of claim 1 further comprising: assessing that a number of times that the physics-based digital model has run in a second computational refinement branch has reached a limit; and halting the second computation refinement branch based on the assessing, such that no candidate input setting is logged for the second computational refinement branch. Claim 7 does not remove or change the abstract limitations identified in claim 1. In particular, the claim requires determining how many times the physics-based digital model has run, comparing that number to a limit, and halting the branch when the limit has been reached. This is still an abstract evaluation and judgment because the claim does not recite any particular technological improvement or specific technique for performing the assessment or halting. As a result, the physics-based digital model is merely a generic computer-implemented tool whose executions are being observed. Therefore, claim 7 only provides more specific recitations of the abstract evaluations and judgments identified in claim 1, and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 7 is not patent eligible.
Regarding claim 8, the claim recites the method of claim 1 wherein the ML model comprises a regression type model. Claim 8 does not remove or change the abstract limitations identified in claim 1. Instead, claim 8 merely specifies that the ML model is a regression type model. A regression model is a mathematical and statistical model used to estimate or predict output values based on relationships between input values and dependent values. Therefore, claim 8 only provides a more specific mathematical model for carrying out the abstract optimization identified in claim 1, and does not recite any particular improvement to the ML model, the computer system, or the industrial equipment. Accordingly, claim 8 does not integrate the judicial exception into a practical application or add significantly more than the judicial exception, and claim 8 is not patent eligible.
Regarding claim 9, the claim recites the method of claim 1 wherein the input setting is selected from the group consisting of a voltage, a pressure regulator setting, and a length of a pipe. Claim 9 does not remove or change the abstract limitations identified in claim 1. Instead, claim 9 merely limits the input setting to examples of industrial process settings, such as a voltage, a pressure regulator setting, or a length of a pipe. These limitations only identify the field or environment in which the abstract optimization is applied and do not recite any particular technological improvement to the equipment, the computer system, the ML model, or the physics-based digital model. Therefore, claim 9 merely limits the abstract idea to a particular field of use (see MPEP § 2106.05(h)), and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 9 is not patent eligible.
Regarding claim 10, the claim recites the method of claim 1 wherein the constrained value is selected from the group consisting of a temperature, a flow rate, a stress or strain level, and a waste emission. Claim 10 does not remove or change the abstract limitations identified in claim 1. Instead, claim 10 merely limits the constrained value to examples of industrial process values, such as temperature, flow rate, stress or strain level, and waste emission. These limitations only identify the type of value being checked in the industrial process environment and do not recite any particular technological improvement to the equipment, the computer system, the ML model, or the physics-based digital model. Therefore, claim 10 merely limits the abstract idea to a particular field of use (see MPEP § 2106.05(h)), and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 10 is not patent eligible.
Regarding claim 11, the claim recites the method of claim 1 wherein the target value is selected from the group consisting of a quantity of product, a chemical purity, an amount of effluent, and a carbon emission. Claim 11 does not remove or change the abstract limitations identified in claim 1. Instead, claim 11 merely limits the target value to examples of industrial process goals, such as a quantity of product, chemical purity, an amount of effluent, and carbon emission. These limitations only identify the type of result or goal being optimized in the industrial process environment and do not recite any particular technological improvement to the equipment, the computer system, the ML model, or the physics-based digital model. Therefore, claim 11 merely limits the abstract idea to a particular field of use (see MPEP § 2106.05(h)), and does not integrate the judicial exception into a practical application or add significantly more than the judicial exception. Accordingly, claim 11 is not patent eligible.
Regarding claims 12-16, the claims recite substantially similar limitations to claims 1-5, respectively, but are directed to a machine-readable tangible medium rather than a method. Therefore, the claims are ineligible under 35 U.S.C 101 for the same reasons above.
Regarding claims 17-21, the claims recite substantially similar limitations to claims 1-5, respectively, but are directed to a computer system rather than a method. Therefore, the claims are ineligible under 35 U.S.C 101 for the same reasons above.
Claim Rejections - 35 USC § 103
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 1-21 are rejected under 35 U.S.C. § 103 as being unpatentable over De Waele et al., U.S. Patent No. US 11,669,063 B2 (hereinafter "Waele") in view of Peralta et al., U.S. Patent Application Publication No. 2007/0168328 A1 (hereinafter "Peralta"), and further in view of Conn et al., U.S. Patent No. US 10,570,717 B2 (hereinafter "Conn")
Regarding Claim 1, Waele teaches a method for optimizing settings in an industrial process (“...the surrogate model can also be applied to other processes and is not limited to refining oil, also described as hydrocarbon processing…This discussion generally relates to tools and methods for analyzing an optimized solution (or solutions) generated from models of hydrocarbon processing systems.”)(e.g., Waele, [Col. 3, lines 40-42 and 48-50])
the method comprising: providing, on a computer system, a physics-based digital model of an industrial process, the digital model having an input setting associated with a piece of equipment in the industrial process, the digital model configured to generate dependent values based on the input setting, the dependent values including a constrained value and a target value (“As briefly mentioned above, each component of the surrogate model system 200, including modeling environment 205, production simulator 220, production data store 290, production data event records 240, and their respective subcomponents, may reside on a computing device (or devices)...”, “The production simulator 220 calculates probable outputs of one or more chemical reactions at specified physical conditions within a production environment. Combining a series of these calculations allows the simulation to show a probable output, given an input. A simulation is constrained by physical conditions in the plant. The production simulator 220 can run thousands of different simulations with different inputs and/or operating conditions. The inputs can be a combination of different oil types…”, and “For example, given an input feed, such as a collection of crude oils, a refinery may produce a series of products using the available equipment. The available equipment may have capacity constraints that are not built into the surrogate model directly. For example, if jet fuel is one output and the capacity of the equipment used to make the jet fuel in a first configuration is 1000 gal/hr, then, in the real world, the plant can produce up to 1000 gal/hr, but can not produce 1200 gal/hr.”, and “In the case of raw material valuation, the objective function f is the profit the decision variable x may include, for example, crude quantities and feedstock quantities. The conditions c can include crude price, product price, compositional data, and unit capacity.”) (e.g., Waele, [Col. 6, lines 5-10 and lines 19-30], [Col. 3, lines 8-15], [Col. 7, lines 13-25]).
Waele's production simulator is interpreted as the claimed physics-based digital model because it executes simulations of the refinery process using input settings, such as crude oil/feedstock quantities, associated with refinery equipment to generate product outputs. The generated product outputs, such as jet fuel output, are interpreted as the claimed dependent values, where the equipment capacity limitation on jet fuel production is interpreted as the claimed constrained value and the profit objective is interpreted as the claimed target value.
providing, on the computer system, a machine learning (ML) model of the industrial process, the ML model having been trained to emulate the physics-based digital model, including the input setting and the dependent values (“As briefly mentioned above, each component of the surrogate model system 200, including modeling environment 205, production simulator 220, production data store 290, production data event records 240, and their respective subcomponents, may reside on a computing device (or devices)...”, “A surrogate model is a machine learned model that uses a collection of inputs and outputs from a simulation of the chemical production process and/or actual production data as training data.”, and “The individual simulation event records within the plurality of simulation event records comprise a description of inputs to a simulation of the chemical production process and a description of outputs from the simulation”) (e.g., Waele, [Col. 6, lines 5-10], [Col. 2, lines 56-60], [Col. 13, lines 6-9]).
generating initial guesses for the input setting (“In order to test the trained model, the input from a production or simulation event record can be provided to the model.”) (e.g., Waele, [Col. 13, lines 64-66]). The Applicant’s disclosure at paragraph 34 states that initial guesses may be estimated based on previous models, experimental data, or real-world settings. Therefore, Waele’s input from a production or simulation event record is interpreted as an initial guess because it is an input value derived from prior simulation or real-world production data and provided to the trained model for evaluation.
maximizing or minimizing the target value produced by the ML model by running the initial guess through the ML model and performing a constrained objective function optimization using the ML model to find a refined guess of the input setting (“At step 540, each input value in the input set is analyzed to confirm it conforms with the constraint. Different constraints may be used. For example, constraints can be used to limit the input space to a maximum and minimum value observed in the training data… A further constraint can be calculated by calculating a hyperplane 320. As a starting point, the hyperplane can be generated by (randomly) selecting a set of crude and product prices, resulting in a regression function that results in the optimization solution 332. If this solution is infeasible when applied to the full exact refinery model, we can add the hyperplane 320 to eliminate the infeasible solution. The process can be repeated until the solution space is suitably narrowed…a solution set is output from the computer-learned surrogate model using the input set as input. The solution set can be the optimal solution of the model for the given input. As mentioned, optimal can be defined as the maximum profit, maximum revenue, maximum generation of a particular finished product, or some other way.” and “a computer-learned surrogate model is trained to generate an optimal statistical output given … production data from the real-world chemical production process … The computer-learned surrogate model comprises a constraint that constrains input variable values to a range bounded by a lowest and a highest input value found in the plurality of simulation event records.”) (e.g., Waele, [Col. 13, lines 15-23 and 30-60]).
logging the refined input as a candidate input setting for the industrial process (“The production data store 290 stores production data for the refinery system. Production data can be any information about the inputs, production, and production outputs of the refinery. In one aspect, the production data forms a series of production data event records 240, such as production data record 240A...”, “The production data store 290 can include any information that describes the refinery inputs, outputs, production setups, and financial data related to any aspect of production”, and “The simulation event records can include data from a computer simulation of a production process, such as those generated by production simulator 220. The simulations can be performed for the express purpose of generating training data, but can also include data from simulations run during optimization exercises or for any other purpose in the course of running simulations”)(e.g., Waele, [Col. 6, lines 31-36 and 66-67], [Col. 7, lines 1-2 and 54-60]). Waele teaches logging because it stores refinery inputs, outputs, production setups, and simulation results in event records, including records from simulations run during optimization exercises. Thus, a refined input generated during optimization and stored with its associated simulated output is interpreted as a logged candidate input setting for the industrial process.
Waele further teaches checking surrogate model results against event records from the production simulator to verify whether the surrogate model output satisfies desired accuracy requirements by comparing simulator output values to surrogate model output values, determining a margin of error, and taking action based on the margin of error. (e.g., Waele, Col. 11, lines 4-21). Waele does not appear to specifically teach selecting an initial guess from the initial guesses, the selected initial guess associated with a computational refinement branch, the branch comprising: inputting the refined guess into the physics-based digital model to check the constrained value; determining, using the physics-based digital model, that the constrained value violates a constraint; subjecting the refined guess and violated constrained value to a constraint correction projection to generate a refined input; inputting the refined input into the physics-based digital model to re-check the constrained value; verifying that the constrained value no longer violates the constraint based on the re-check; selecting among multiple candidate input settings for a best input setting, each candidate input setting originating from a different initial guess of the initial guesses and a respective computational refinement branch.
However, Peralta teaches selecting an initial guess from the initial guesses, the selected initial guess associated with a computational refinement branch (“ISTO includes multiple cycles. Each cycle includes the steps: (1) initializing ISTO with a solution strategy 120, 930; (2) defining the multi-dimensional space tube (subspace) 130, 920; (3) generating simulations (used for creating or training surrogate simulator) …”, “To avoid exploring the entire solution space, ISTO generates new strategies within and near the leading portion (head) of the space tube. The space tube 410 is defined around the initial strategy (first cycle) or best strategy to date (subsequent cycles).”, “Parallel processing facilitates employing multiple space tubes simultaneously”)(e.g., Peralta, paragraphs [0035] and [0073]-[0074]). Peralta’s generated strategies are interpreted as the initial guesses, the initial strategy or best strategy to date is interpreted as the selected initial guess, and the space tube in which Peralta performs simulation and surrogate optimization is interpreted as the computational refinement branch associated with the selected initial guess.
inputting the refined guess into the physics-based digital model to check the constrained value and determining, using the physics-based digital model, that the constrained value violates a constraint (“ISTO includes multiple cycles. Each cycle includes the steps: ...(5) performing optimization that uses the surrogate simulators 160, 950; (6) analyzing the optimal strategy, computed in step 5, concerning objective function value and constraint violations 190, 210, 960; (7) evaluating whether space tube radius modification is required”, Figure 7 step 6 states "The optimal strategy computed by the SS is simulated using the original simulator to determine whether or not it is feasible. The strategy is accepted if all constraints are satisfied and OF is equal to or better than the best OF to date; otherwise the strategy is rejected”)(e.g., Peralta, paragraph [0073] and figure 7). The original simulator is interpreted as the claimed physics-based digital model because Peralta describes the original simulator as the accurate simulator used to simulate system response and determine feasibility (Peralta, paragraph 7-8).
inputting the refined input into the physics-based digital model to re-check the constrained value and verifying that the constrained value no longer violates the constraint based on the re-check (“ISTO includes multiple cycles. Each cycle includes the steps: … performing optimization that uses the surrogate simulators 160, 950; (6) analyzing the optimal strategy, computed in step 5, concerning objective function value and constraint violations 190, 210, 960; (7) evaluating whether space tube radius modification is required 220, 240, 970; and (8) evaluating stopping criteria 230, 980. Step 9, optimization using the original simulator(s) 260, 990, refines the optimal strategy developed during the previous steps.”, “In Step 7 970 ISTO evaluates whether the space tube radius should be modified for the next cycle, and makes the modification 220, 240. ISTO can employ any desired modification method. For illustration here, the space tube radius is reduced if, in Step 6, the optimal strategy is rejected 220. If the optimal strategy is accepted in Step 6, but was rejected in the previous cycle, the radius is increased 240”, and “In Step 9, ISTO switches to using original simulators and can change optimizers to refine the optimal strategy resulting from Steps 2-7 cycling. Step 9 990 is valuable if constraints are so tight that surrogate simulator accuracy becomes inadequate. Step 9 is continued until a stopping criterion is satisfied 270”) (e.g., Peralta, paragraphs [0073], [0077], and [0080]).
Peralta teaches a repeated refinement workflow in which a strategy generated using surrogate simulators is checked for constraint violations. When the strategy is rejected, Peralta modifies the space tube radius for the next cycle, thereby changing the search space used to generate a further strategy. Peralta also teaches that a later strategy may be accepted after a previous rejection, which indicates that the newly generated strategy is re-evaluated and found to satisfy the applicable constraints. Thus, the strategy generated after the rejection and tube modification is interpreted as the refined input. Peralta further teaches that, after the repeated cycling in phase 1, the system switches to the original simulator, rather than the surrogate simulator, to refine the optimal strategy when constraints are tight and predictive accuracy is important. Accordingly, Peralta’s later use of the original simulator to refine and evaluate the strategy is interpreted as inputting the refined input into the physics-based digital model to re-check the constrained value, and the later acceptance of the strategy after a prior rejection is interpreted as verifying that the constrained value no longer violates the constraint based on the re-check.
selecting among multiple candidate input settings for a best input setting, each candidate input setting originating from a different initial guess of the initial guesses and a respective computational refinement branch (“The above descriptions of prediction, interpolation, or extrapolation methods, including preferred embodiments contained herein, are to be construed as merely illustrative and not a limitation of the scope of the present invention in any way. It will be obvious to those having skill in the art that many changes may be made to the details of the above-described embodiments without departing from the underlying principles of the invention…Anything which functions the same as, or equivalently to, a means for prediction, interpolation, or extrapolation falls within the scope of this element. Evolutionary algorithms (EAs) are search methods that utilize a form of natural selection and survival of the fittest. EAs differ from more traditional optimization techniques in that they involve a search from a “population” of solutions, not from a single point. Each iteration of an EA involves a competitive selection that weeds out poor solutions. The solutions with high “fitness” are “recombined” with other solutions by swapping parts of a solution with another…EAs are often viewed as global optimization methods although convergence to a global optimum is only guaranteed in a weak probabilistic sense. However, an EA strength is that they perform well on “noisy” functions where there may be multiple local optima. EAs tend not to get “stuck” on local minima and can often find globally optimal solutions.”) (e.g., Peralta, paragraphs [0051]- [0052], and [0060]-[0061]).
Peralta’s population of solutions is interpreted as the claimed multiple candidate input settings because each solution represents a possible strategy or set of decision variables being evaluated by the optimizer. Peralta’s initialized population is interpreted as the initial guesses because the population provides multiple starting solution candidates for the optimization search. Peralta’s competitive selection is interpreted as selecting among the multiple candidate input settings because the algorithm compares the candidate solutions and weeds out poorer solutions. Peralta’s high-fitness solutions are interpreted as better candidate input settings because they are the solutions retained and used to generate subsequent solutions. Peralta’s teaching that evolutionary algorithms avoid local minima and can often find globally optimal solutions further supports interpreting the process as selecting a best input setting from among multiple candidate input settings, rather than simply refining a single candidate.
Neither Waele nor Peralta appear to specifically teach subjecting the refined guess and violated constrained value to a constraint correction projection to generate a refined input.
However, Conn teaches subjecting the refined guess and violated constrained value to a constraint correction projection to generate a refined input (“Constraints in derivative free optimization may be handled using an exact penalty function. For example, in maximizing emulsion rate, a weighted combination of the constraint violations may be added to the objective function so as the weights increase the result is encouraged to be feasible.” and “The relaxed feasible solution proposition seeks to find the approximate optimal value of the surrogate model while accounting for the feasible region as defined by the constraints on the SAGD system 104, using techniques such as projections, active set, augmented penalty functions, etc. The relaxed feasible solution proposition itself may involve a number of iterations.”) (e.g., Conn, [Col. 16, lines 51-56], [Col. 18, lines 16-22]).
Conn teaches applying a correction to a constraint violating optimization result because Conn incorporates weighted constraint violations into the objective function to encourage feasibility and then determines an approximate optimal value while accounting for the feasible region using projections, active set techniques, and augmented penalty functions. Since the feasible region is defined by the SAGD constraints, Conn’s projection optimization uses the violated constraint information to adjust the control-parameter value toward a feasible value. Accordingly, Conn’s constraint violating control parameter value is interpreted as the refined guess, Conn’s constraint violation is interpreted as the violated constrained value, and Conn’s projection/feasible region correction is interpreted as generating a corrected control parameter value corresponding to the claimed refined input.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Waele, Peralta, and Conn before him or her, to improve Waele’s surrogate model system for optimizing chemical processing input settings subject to physical plant constraints. The improvement would have included using Peralta’s simulator refinement workflow to check surrogate generated candidate settings and using Conn’s projection based correction techniques to adjust constraint violating candidate settings toward feasible values.
A person of ordinary skill in the art would have looked to Peralta and Conn when improving Waele because the references address the same type of constrained optimization problem. Waele is directed to optimizing hydrocarbon processing inputs using a production simulator and a machine learned surrogate model. Conn is similarly directed to optimizing oil well control parameters, including steam rates and pressures, while satisfying operational constraints. Peralta is also relevant because it teaches reducing computation in simulation based optimization while still using an accurate “original” simulator to check objective value, feasibility, and constraint violations. Peralta further teaches optimization of extraction well strategies, including pumping or tubing related parameters. Therefore, the references are analogous because they all concern improving computer implemented optimization of process settings while maintaining physical or operational constraints.
A person of ordinary skill in the art would have been motivated to incorporate Peralta’s refinement workflow into Waele because Peralta expressly teaches advantages directed to conserving computing resources and improving execution speed in simulation based optimization (e.g. paragraph [0015]). Peralta teaches that ISTO is useful when predicting system response is computationally intensive and many predictions are needed during optimization. Peralta further teaches a scalable framework that may use multiple simulators and multiple space tubes, including parallel processing, to evaluate different candidate strategies efficiently (e.g. paragraph [0020]). Peralta also teaches checking a surrogate generated strategy with the original simulator and rejecting the strategy when it is infeasible or fails to improve the objective value (e.g. paragraph [0073] and figure 7). Peralta’s use of feasibility checks, stopping criteria, and simulator based refinement would have predictably improved Waele by reducing unnecessary processing of unsuitable candidate settings while preserving the accuracy of the production simulator when physical constraints are important.
A person of ordinary skill in the art would have been further motivated to incorporate Conn’s projection based correction techniques because the combined Waele and Peralta workflow identifies when a surrogate generated candidate is infeasible, while Conn teaches a known way to use that violation information. The combined Waele and Peralta workflow teaches checking a candidate setting with a simulator and rejecting the candidate when the constraints are not satisfied. Conn teaches handling constraint violations by using weighted constraint violations, projections, active set techniques, and augmented penalty functions to account for the feasible region. Therefore, adding Conn would predictably improve the system by converting a rejected or infeasible candidate into a corrected candidate for further evaluation, rather than simply abandoning the candidate and continuing the search. This would reduce computational waste and help the system achieve an optimal feasible solution faster.
Regarding Claim 2, Conn teaches the method of claim 1 wherein the constrained function optimization uses a gradient-based algorithm or a trust region method. The examiner is selecting a trust region method from the provided options. (“In some embodiments, given a suitable predictive model and simulator of the SAGD system 104 the optimization may be carried out using a trust region method.”)(e.g., Conn, [Col. 11, lines 20-22]).
Regarding Claim 3, Peralta teaches the method of claim 1 wherein the constraint correction projection uses an input parameter perturbation technique (“ISTO can also employ any sort of optimization algorithm. Examples are classical operations research techniques (such as gradient search, outer approximation, or other methods), heuristic methods (such as genetic algorithm (GA), simulated annealing (SA), tabu search (TS), hybrids, etc.), decision-tree, computational intelligence, or other techniques.”, “The original SA scheme was that an initial state of a thermodynamic system was chosen at energy E and temperature T, holding T constant the initial configuration is perturbed and the change in energy dE is computed… The current state of the thermodynamic system is analogous to the current solution to the combinatorial problem, the energy equation for the thermodynamic system is analogous to the objective function, and the ground state is analogous to the global minimum.”) (e.g., Peralta, paragraphs [0019], and [0068]-[0069]). Peralta teaches that ISTO may use simulated annealing as an optimization technique. While describing simulated annealing, Peralta explains that the current configuration is perturbed and that the current configuration corresponds to the current solution of the optimization problem. Therefore, Peralta’s perturbation of the current solution is interpreted as perturbing the current input parameter set to generate a new candidate input.
Regarding Claim 4, Conn teaches the method of claim 1 wherein the constraint correction projection includes: determining an amount of violation of the constraint (“Constraints in derivative free optimization may be handled using an exact penalty function. For example, in maximizing emulsion rate, a weighted combination of the constraint violations may be added to the objective function so as the weights increase the result is encouraged to be feasible.”)(e.g., Conn, [Col. 16, lines 51-56]). Conn’s weighted combination of constraint violations is interpreted as determining the amount of violation of the constraint because the penalty term depends on the extent to which the constraint is violated. In other words, Conn must identify the violation amount in order to weight the constraint violation and add it to the objective function to produce an improved result.
determining a targeted constraint output for the ML model and computing a gradient of the ML model at the refined guess for the constrained value and the target value (“The simulator or black box optimizer of the SAGD system 104 may be used in conjunction with the predictive model or surrogate to optimize, improve or otherwise meet one or more objectives subject to one or more constraints. Objectives may include but are not limited to increasing or maximizing emulsion production, decreasing or minimizing a steam to oil ratio, increasing or maximizing a net present value, etc.” and “the surrogate model is formed based on curvature information obtained from the predictive model of the SAGD system 104 when the predictive model of the SAGD system 104 is differentiable with respect to the control parameters. The curvature information may be obtained from the predictive model of the SAGD system 104 using, for example, low rank gradient based estimates for Hessians incorporated into quasi-Newton like methods…”) (e.g., Conn, [Col. 14, lines 56-63] and [Col. 17, lines 48-57]).
Conn teaches these limitations because the predictive model is used to optimize model outputs, such as emulsion production or steam-to-oil ratio, while satisfying operational constraints, thereby identifying a model output that satisfies the constraint. Conn further teaches computing gradient information at the refined control value because the surrogate model is formed from curvature information of the differentiable predictive model using low-rank gradient-based Hessian estimates and quasi-Newton methods, which provide sensitivity information for how the objective value and constrained value change with respect to the control parameters.
determining a vector direction of the gradient so as to minimally affect the target value, computing a direction of correction for the constrained value, and updating the refined guess based on the computed direction of correction (“the surrogate model is formed based on curvature information obtained from the predictive model... using, for example, low rank gradient based estimates for Hessians incorporated into quasi-Newton like methods..” and “Relaxed feasible solution proposition, in some embodiments, involves finding the approximate optimal value of the surrogate model within a specified level of confidence in a trust region neighborhood of the current best surrogate model input parameters … the relaxed feasible solution proposition seeks to find the approximate optimal value of the surrogate model while accounting for the feasible region as defined by the constraints... using techniques such as projections, active set, augmented penalty functions, etc.”)(e.g., Conn, [Col. 17, lines 49-63] and [Col. 18, lines 8-22]).
Conn teaches these limitations because Conn uses gradient-based curvature information to determine how the current best surrogate input parameters should be moved within a trust-region neighborhood. Conn is not simply fixing the constraint without regard to the objective. Instead, Conn seeks an approximate optimal value of the surrogate model while also accounting for the feasible region defined by the constraints. Therefore, the movement from the current surrogate input parameters is interpreted as a correction direction that reduces or avoids the constraint violation without unnecessarily sacrificing the objective value. The resulting feasible trust-region update corresponds to updating the refined guess based on the computed direction of correction.
Regarding Claim 5, Conn teaches the method of claim 4 wherein the constraint correction projection is repeated in a loop (“The relaxed feasible solution proposition seeks to find the approximate optimal value of the surrogate model while accounting for the feasible region as defined by the constraints on the SAGD system 104, using techniques such as projections, active set, augmented penalty functions, etc. The relaxed feasible solution proposition itself may involve a number of iterations.” and “the above-described procedure for searching the control space may be repeated until convergence upon a feasible solution of the mixed integer non-linear problem is achieved.” )(e.g., Conn, [Col. 18, lines 16-22 and 45-48]).
Regarding Claim 6, Peralta teaches the method of claim 1 further comprising: assessing that a number of times that the physics-based digital model has run is fewer than a limit; and allowing the inputting of the refined input into the physics-based digital model based on the assessing (“ISTO includes multiple cycles. Each cycle includes the steps … evaluating whether space tube radius modification is required 220, 240, 970; and (8) evaluating stopping criteria 230, 980…”, “Step 8 980 determines whether ISTO continues using surrogate simulators in optimization, switches to the original simulators, or halts…Step 9 is continued until a stopping criterion is satisfied 270 (an example is completion of a specified number of simulations).”)(e.g., Peralta, paragraphs [0073], [0080], and figure 7). Peralta’s specified number of simulations is interpreted as the claimed limit on the number of times the physics-based digital model is run. Peralta’s stopping criteria evaluation is interpreted as assessing whether that limit has been reached. Because Peralta teaches that the process continues until the stopping criterion is satisfied, the process is allowed to continue when the number of simulations is fewer than the specified limit. Therefore, Peralta’s continued use of the original simulator before the simulation limit is reached is interpreted as allowing the refined input to be input into the physics-based digital model based on the assessing.
Regarding Claim 7, Peralta teaches the method of claim 1 further comprising: assessing that a number of times that the physics-based digital model has run in a second computational refinement branch has reached a limit; and halting the second computation refinement branch based on the assessing, such that no candidate input setting is logged for the second computational refinement branch (“ISTO includes multiple cycles. Each cycle includes the steps … evaluating whether space tube radius modification is required 220, 240, 970; and (8) evaluating stopping criteria 230, 980…”, “Step 8 980 determines whether ISTO continues using surrogate simulators in optimization, switches to the original simulators, or halts…Step 9 is continued until a stopping criterion is satisfied 270 (an example is completion of a specified number of simulations).”)(e.g., Peralta, paragraphs [0073], [0080], and figure 7). Peralta’s specified number of simulations is interpreted as the limit on the number of times the physics-based digital model is run in that branch. Peralta’s stopping criterion is interpreted as assessing that the limit has been reached and halting the branch. Because the branch is halted upon reaching the limit, no accepted strategy is produced from that branch, and therefore no candidate input setting is logged for the second computational refinement branch.
Regarding Claim 8, Waele teaches the method of claim 1 wherein the ML model comprises a regression type model (“Various machine learning models may be used to implement the surrogate model. Implementations using a regression surrogate, neural network, and convex hull models are described below. ”)(e.g., Waele, [Col. 7, lines 10-12]).
Regarding Claim 9, Conn teaches the method of claim 1 wherein the input setting is selected from the group consisting of a voltage, a pressure regulator setting, and a length of a pipe. The examiner is selecting a pressure regulator setting from the provided options. (“Model inputs may include, by way of example, heel steam rate, toe steam rate, cumulative injected steam and energy or integrated total mass and total energy inserted (as measured from some defined start time), gas casing pressure, emulsion pressure, mass differential, cumulative mass differential, energy differential and cumulative energy differential.”)(e.g., Conn, [Col. 13, lines 24-29]). Conn’s gas casing pressure and emulsion pressure are interpreted as pressure regulator settings because Conn identifies these pressure values as controls or model inputs used to operate and optimize the SAGD industrial process. Under a broadest reasonable interpretation, a pressure regulator setting includes a pressure-related control value that regulates, sets, or controls pressure in the industrial process. Therefore, Conn’s pressure inputs satisfy the claimed limitations.
Regarding Claim 10, Conn teaches the method of claim 1 wherein the constrained value is selected from the group consisting of a temperature, a flow rate, a stress or strain level, and a waste emission. The examiner is selecting a temperature from the provided options. (“Examples of objective functions include but are not limited to maximization of emulsion flow rate, minimizing steam to oil ratio, maximizing net present value, etc. Constraints may be related to the controls, such as maximum heel or toe steam rate, maximum emulsion pressure, etc. Some constraints may be related to the state of the SAGD system 104, such as maximum surface pressure, maximum bottom hole pressure, minimum sub-cool temperature, maximum variations of temperature across different zones, etc.”)(e.g., Conn, [Col. 11, lines 52-61]).
Regarding Claim 11, Waele teaches the method of claim 1 wherein the target value is selected from the group consisting of a quantity of product, a chemical purity, an amount of effluent, and a carbon emission. The examiner is selecting a quantity of product from the provided options. (“A surrogate model can be used both for predicting revenue/profit ($) and product quantities (barrel/day)...”)(e.g., Waele, [Col. 3, lines 24-25]).
Regarding Claim 12-16, the claims recite substantially similar limitations to claim 1-5, but add a machine-readable tangible medium embodying information indicative of instructions for causing one or more machines to perform operations for optimizing input parameters for an industrial process ("The technology described herein may be described in the general context of computer code or machine-useable instructions, including computer-executable instructions such as program components, being executed by a computer or other machine, such as a personal data assistant or other handheld device. Generally, program components, including routines, programs, objects, components, data structures, and the like, refer to code that performs particular tasks or implements particular abstract data types. The technology described herein may be practiced in a variety of system configurations, including handheld devices, consumer electronics, general-purpose computers, specialty computing devices, etc. Aspects of the technology described herein may also be practiced in distributed computing environments where tasks are performed by remote-processing devices that are linked through a communications network.") (e.g., Waele, [Col. 17, lines 64-67] and [Col. 18, lines 1-12]).
Therefore, claims 12-16 are rejected under 35 U.S.C. § 103 for the same reasons set forth above with respect to claims 1-5.
Regarding Claim 17-21, the claims recite substantially similar limitations to claims 1-5, but add a system for optimizing input parameters for an industrial process, the system comprising: a memory; and at least one processor operatively coupled with the memory and executing program code from the memory (“Referring to the drawings in general, and initially to FIG. 8 in particular, an exemplary operating environment for implementing aspects of the technology described herein is shown and designated generally as computing device 800…With continued reference to FIG. 8, computing device 800 includes a bus 810 that directly or indirectly couples the following devices: memory 812, one or more processors 814, one or more presentation components 816, input/output (I/O) ports 818, I/O components 820, and an illustrative power supply 822…”)(e.g., Waele, [Col. 17, lines 54-57] and [Col. 18, lines 13-18]).
Therefore, claims 17-21 are rejected under 35 U.S.C. § 103 for the same reasons set forth above with respect to claims 1-5.
Conclusion
The prior art made of record, listed on PTO-892, and not relied upon is considered pertinent to applicant's disclosure.
Phan et al., U.S. Patent Application Publication No. 2023/0297073 A1, teaches optimizing industrial processes using machine learning. Phan teaches learning relationships between process inputs, set-points, and outputs, deriving regression functions from historical data, merging the regression functions into a unified optimization problem, and determining optimal set-points for operating the industrial system. This reference is pertinent to applicant’s disclosure because it relates to using machine learning and optimization to determine operating settings for industrial processes.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AREEBAH FATIMA whose telephone number is (571)270-0294. The examiner can normally be reached 9am - 5pm.
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, Rehana Perveen can be reached at (571) 272-3676. 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.
/AREEBAH FATIMA/ Examiner, Art Unit 2189
/REHANA PERVEEN/ Supervisory Patent Examiner, Art Unit 2189
Prosecution Insights