DETAILED ACTION
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 .
Response to Amendment/Argument
1. Applicant’s amendment to paragraph [0158] of the specification, submitted to correct a translation error, has been acknowledged.
2. Amendment to claim 4 overcomes the previous rejections under 35 U.S.C. 112(b) with respect to claims 4, 5, 7, and 9. The rejections of claims 5, 8, 10, and 11 under 35 U.S.C. 112(b) is moot in view of cancellation of these claims.
3. Applicant’s arguments to the rejection under 35 U.S.C. 101 filed on December 10, 2025 have been fully considered but they are not persuasive. Applicant asserts that the claims recite technological improvements directed to improving accuracy of modeling to predict energy usage patterns and cites paragraphs [0006]-[0008] and [0132] of the specification as support. However, improving the accuracy of predictions or data modeling does not, by itself, demonstrate an improvement to computer functionality or another technology. Instead, the claims recite analyzing data, comparing models, clustering models based on similarity, and updating models using mathematical relationships. Improving prediction accuracy through data analysis and mathematical optimization constitutes an improvement in abstract information processing rather than an improvement to computer technology itself.
Applicant further argues that the amended claims integrate the abstract idea into a practical application because they recite determining whether a federated model exists, comparing gradients of models, and clustering energy prediction models into the federated model. These additional limitations do not integrate the abstract idea into a practical application. The limitations describe evaluating data relationships between models and performing mathematical comparison operations.
In particular, Applicant emphasizes that the claims compare gradient models. A gradient represents a derivative or rate of change of model parameters and is a fundamental mathematical construct. Comparing gradients therefore represents comparison of mathematical values and relationships. Whether similarity is determined through general similarity metrics or through gradient comparison, the claims still recite mathematical analysis of data relationships. Such evaluation of mathematical relationships corresponds to mathematical concepts and mental evaluation processes that can be performed in the human mind, using basic computational tools such as pen and paper and/or calculator.
The recitation of determining whether a federated model exists in memory does not change the character of the claims. The limitation merely recites checking for stored data before performing mathematical comparison and clustering operations. Such memory access and data retrieval operations are generic computer functions and constitute insignificant extra-solution activity. Similarly, the recitation of clustering models into a federated model and updating the federated model describes organizing and updating stored information based on mathematical comparison results and does not reflect an improvement to computer architecture, memory structure, or network communication.
Applicant also relies on disclosure describing improving service performance and quality by using models from multiple households within the same region and updating a shared model based on similarity. This argument is unpersuasive. The claims do not recite any specific technological improvement to computer architecture, memory structure, communications, or ESS hardware. Instead, the claims recite comparing model information (e.g., gradient/vectors), determining similarity, clustering models based on that similarity, and updating a stored model. These steps address statistical/modeling objectives (e.g., improving prediction accuracy and reducing over-fitting) through mathematical analysis and aggregation of model information, which represents an abstract improvement in mathematical processing of data rather than improvement to computer functionality or another technology.
With respect to Step 2B, the claims do not recite additional elements that amount to significantly more than the abstract idea. The processor and memory recited in the claims are described at a high level of generality and perform routine computer functions including receiving data, storing data, retrieving stored data, comparing data, and updating stored data. The claims do not recite specialized hardware, improve data structures, improved distributed training architecture, or unconventional computing techniques. Further, techniques such as gradient calculation, gradient comparison, clustering models based on similarity, and applying optimization algorithms such as PCGrad were know practices in machine learning and therefore represent well-understood, routine, and conventional activity within the field.
Accordingly, the amended claims remained directed to judicial exceptions without significantly more, and the rejection under 35 U.S.C. 101 is maintained.
4. Applicant’s arguments regarding the rejection under 35 U.S.C. 103 are not persuasive. As noted in the Office Action, the rejection of independent claim 1 and 15 are now maintained under 35 U.S.C. 103 rather than 102. Although Applicant presents arguments asserting Khan does not anticipate the claims, anticipation is no longer an issue because the present rejection is based on obviousness over Khan in view of Obrenović and/or Liu.
Applicant’s argument regarding Khan and the proposed combinations have been fully considered but are not persuasive. Under the broadest reasonable interpretation, Khan’s consumption profiles and the model representations derived from them correspond to energy prediction models, and Khan’s aggregated forecasting models and cluster centroids correspond to a federated model because they are global models derived from multiple local inputs. Accordingly, determining similarity between a model-related representation and an aggregated model representation in Khan corresponds to the claimed determination of similarity between an energy prediction model and a federated model.
Applicant’s contention that Khan merely groups consumption profiles and does not determine whether a particular model is similar to a federated model is not persuasive. Khan employs hybrid distance metrics to compare model-related representations and assign them to the closest aggregated model (cluster centroid). This constitutes determining similarity under BRI, regardless of whether the underlying representation is described as a consumption profile or an energy prediction model.
Applicant further argues that Khan does not disclose comparing gradients. However, the claims do not require any particular gradient-descent technique or a specific form of gradient representation. Gradients are mathematical vector representations of model behavior, and Obrenović teaches comparing multi-dimensional vector representations using cosine similarity, Euclidean distance, Manhattan distance, and related metrics. These teachings correspond to the broadly recited comparison of gradients under BRI. Khan likewise compares vector-based model representations using hybrid distance metrics.
Applicant also asserts that Khan does not disclose clustering energy prediction models into a federated model or updating such a model. Khan groups multiple model-related representations into clusters and generates aggregated models from those grouped inputs, and updates these aggregated models from those grouped inputs, and updates these aggregated models when new or updated profiles are assigned to a cluster. Under BRI, this corresponds to clustering energy prediction models into, and updating, a federated model to improve prediction accuracy.
Accordingly, although Applicant argues that Khan alone does not teach the amended features, the current 103 rejection includes Obrenović, and the combination teaches or suggests the limitations claimed.
Applicant also argues that neither Obrenović no Liu discloses or suggests the amended features. This argument is not persuasive. As discussed above, Obrenović teaches comparing multi-dimensional vector representations using cosine similarity, Euclidean distance, Manhattan distance, and related metrics. Under the broadest reasonable interpretation, these disclosures correspond to the claimed comparison of gradients, which are mathematical vector representations of model behavior. Accordingly, Obrenović provides the teachings relied upon in the current 103 rejection for broadly recited gradient-based similarity determination.
Applicant further contends that Liu does not disclose determining whether an energy prediction model is similar to a federated model by comparing gradients. However, Liu is cited for teaching gradient manipulation techniques, including modifying, projecting, and comparing gradient vectors in multi-task learning. These disclosures demonstrate that gradient-based comparison and manipulation were well-known in the art, and they reinforce the obviousness of applying vector-based similarity techniques, such as those taught by Obrenović, to model-related representations as in Khan. Liu is not relied upon to disclose a federated model, but rather to show that gradient-based comparison techniques were known and would have been obvious to apply in the context of Khan’s aggregated models.
Accordingly, Applicant’s arguments regarding Obrenović and Liu do not overcome the current 103 rejection. The combination of Khan with Obrenović, and further in view of Liu where cited, teaches or suggests the limitations as claimed.
Applicant’s arguments regarding independent claim 15 are not persuasive. Claim 15 recites limitations that correspond to those addressed above with respect to independent claim 1. For the same reasons, claim 15 is not anticipated by Khan and is unpatentable over Khan in view of Obrenović and/or Liu.
Applicant further asserts that dependent claims 2-4, 6, 7, 9 and 12-14 are patentable by virtue of their dependency. However, because the independent claims from which they depend are unpatentable for the reasons discussed above, the dependent claims likewise are unpatentable.
For the foregoing reasons, Applicant’s arguments have been fully considered but are not persuasive. The combination of Khan with Obrenović, and further in view of Liu where cited, continues to teach or suggest the limitations as claimed. Accordingly, the rejection of the claims under 35 U.S.C. 103 is maintained.
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 1-4, 6-7, 9, and 12-15 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more.
101 Subject Matter Eligibility Analysis
Step 1: Claims 1-4, 6-7, 9, and 12-15 are within the four statutory categories (a process, machine, manufacture or composition of matter).
Step 2A Prong One, Step 2A Prong Two, and Step 2B Analysis:
Step 2A Prong One asks if the claim recites a judicial exception (abstract idea, law of nature, or natural phenomenon). If the claim recites a judicial exception, analysis proceeds to Step 2A Prong Two, which asks if the claim recites additional elements that integrate the abstract idea into a practical application. If the claim does not integrate the judicial exception, analysis proceeds to Step 2B, which asks if the claim amounts to significantly more than the judicial exception. If the claim does not amount to significantly more than the judicial exception, the claim is not eligible subject matter under 35 U.S.C. 101.
None of the claims represent an improvement to technology.
Claims 1-4, 6-7, 9, and 12-14 are directed to an apparatus with a memory and processor which is a machine. Claim 15 is directed to a method consisting of a series of steps, meaning that it is directed to the statutory category of process.
Regarding claim 1, the following claim elements are abstract ideas:
determine whether a federated model exists in the memory for determining whether each of the energy prediction models is similar to the federated model (This is an abstract idea of a “mental process.” Determining whether a model exists requires observation and judgement. A model be represented as a record of household energy features, such as size of the house, appliance types, or usage history. A person could review records for a given area, observe whether another record with similar features exists, and judge the similarity between the two records. Such observation and judgement can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas. See MPEP 2106.04(a)(2)(III).);
based on determining that the federated model exists in memory, for each of the energy prediction model (This is an abstract idea of a mental process. The limitation recites a determination that can be performed in the human mind by evaluating the presence or absence of information.):
determine whether the energy prediction model is similar to the federated model by comparing gradients of the energy prediction model and gradients of the federated model (This is an abstract idea of a “mental process.” Determining similarity between models requires observation and judgement. Each model may be represented as a record of household energy features, such as a size of the house, appliance types, or usage history. After determining that a record with such features exists in a given area, a person could review the household’s own record of features alongside that record and, based on observation and judgement, determine the degree of similarity between the two. Such observation and judgement can be performed in the human mind, or with pen and paper and thus falls within the mental process grouping of abstract ideas.);
cluster the energy prediction models that are determined to be similar to the federated model into the federated model to update the federated model, in order to improve the accuracy of the modeling to predict energy usage patterns (This is an abstract idea of a “mental process.” Clustering models requires observation and judgement. Each model may be represented as a record of household energy features, such as size of the house, appliance types, or usage history. A person could review a household’s record, decide whether it belongs to a particular group of similar records. This type of classification – assigning records to a group according to observed similarity – is a cognitive task that can be performed mentally or with pen and paper, and thus falls within the mental process group of abstract ideas.)
The following claim elements are additional elements which, taken alone or in combination with the other elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
a memory and a processor (This is a high-level recitation of generic computer components for performing the abstract idea. See MPEP 2106.05.)
receive a plurality of energy prediction models, each of the energy prediction models being for a respective household of a plurality of households (This limitation amounts to adding insignificant extra-solution activity to the judicial exception. Receiving predictions models (i.e., mere data gathering in conjunction with the abstract idea) is directed to a well understood, routine, and conventional activity of data transmission see MPEP 2106.05(d)(II)(i).);
Regarding claim 2, the rejection of claim 1 is incorporated herein. Further, claim 2 recites the following additional elements, which taken alone or in combination with other elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
receive the plurality of energy prediction models from an energy storage system (ESS) of each household located within a same area (This limitation amounts to adding insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g). Receiving a plurality of models (i.e., mere data gathering in conjunction with an abstract idea) is directed to a well understood routine conventional activity of data transmission see MPEP 2106.05(d)(II)(i).),
Regarding claim 3, the rejection of claim 1 is incorporated herein. Further, claim 3 recites the following abstract ideas:
based on determining that the federated model does not exist in the memory, generate a new federated model to be stored in the memory based on an input energy prediction model (This is an abstract idea of a “mental process.” Generating a new model based on the absence of an existing one requires observation and judgement. A model may be represented as a record of household energy features, such as size of the house, appliance types, or usage history. A person could first review records in a given area and, based on observation and judgement, determine that no such record exists. The person could then create a new record using an input household’s features and decide to store it for later comparison. These steps rely on observation and judgement that can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas.).
Regarding claim 4, the rejection of claim 1 is incorporated herein. Further, claim 4 recites the following abstract ideas:
based on determining that the federated model exists in the memory, determine whether one or more federated models are stored in the memory (This is abstract idea of a “mental process.” Determining whether records exist requires observation and judgement. A model may be represented as a record of household energy features, such as size of the house, appliance types, or usage history. After first observing that a record for such a model exists, a person could then review the same location to check whether one or more additional records are stored there. These steps of observing records and judging whether multiple record exist can be performed mentally or with pen and paper.);
based on determining that there is one federated model stored in the memory, compare an input energy prediction model and the one federated model (This is an abstract idea of a “mental process.” Determining similarity between models requires observation and judgement. Each model may be represented as a record of household energy features, such as size of the house, appliance types, or usage history. After first observing that a single record exists in a given area, a person could then compare the input household’s record of features with that record and, based on observation and judgement, determine the degree of similarity between them.).
Regarding claim 6, the rejection of claim 4 is incorporated herein. Further, claim 6 recites the following abstract ideas:
based on determining that the input energy prediction model and the one federated model are different from each other, generate a new federated model to be stored in the memory, based on the input energy prediction model (This is an abstract idea of a “mental process.” Determining differences and deciding to create a new model requires observation and judgement. Each model may be represented as a record of household energy features, such as size of the house, appliance types, or usage history. A person could review the household’s record alongside another record, observe the different features and note it for future reference. Such observation and judgement can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas.).
Regarding claim 7, the rejection of claim 4 is incorporated herein. Further, claim 7 recites the following abstract ideas:
based on determining that a plurality of federated models are stored in memory, compare the input energy prediction model and each of the plurality of federated models (This is an abstract idea of “mental process.” Determining similarity requires observation and judgement. Here, the “federated models” can be understood as groupings of similar records of household energy features, such as size of a house, appliance types, or usage history. A person could review the input record of features alongside each grouping of records and, based on observation and judgement, determine how similar the features are. Such evaluation of multiple records through comparison and judgement can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas. See MPEP 2106.04(a)(2)(III).) ;
select a federated model of the plurality of federated models having a highest similarity with respect to the input energy prediction model (This is an abstract idea of a “mental process.” Selecting among groupings requires observation and judgement. The “federated models” can be understood as groupings of similar records of household energy features, such as size of the house, appliance types, or usage history. A person could compare the input record of features to each grouping of records, judge which grouping is most similar, and select that grouping. Such evaluation and selection through observation and judgement can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas. See MPEP 2106.04(a)(2)(III).).
Regarding claim 9, the rejection of claim 7 is incorporated herein. Further, claim 9 recites the following abstract ideas:
based on determining that the input energy prediction model and all of the plurality of federated models are different from each other, generate a new federated model to be stored in the memory, based on the input energy prediction model (This is an abstract idea of a “mental process.” Determining differences and deciding to create a new grouping require observation and judgement. Each model may be represented as a record of household energy features, such as size of the house, appliance types, or usage history. A person could review the input record alongside each grouping of records, observe that the features are different from all of them, and, using judgement, decide to create a new grouping from the input record and note it for later reference. Such observation and judgement can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas.).
Regarding claim 12, the rejection of claim 1 is incorporated herein. Further, claim 12 recites the following additional elements, which taken alone or in combination with other elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
update the federated model based on a project conflicting gradients (PCGrad) algorithm (The step of “updating” a model is merely a generic data operation that amounts to storing and retrieving information in memory, which has been recognized by the courts as well-understood, routine, and conventional activity. See MPEP 2106.05(d)(II)(iv).).
Regarding claim 13, the rejection of claim 1 is incorporated herein. Further, claim 13 recites the following additional elements, which taken alone or in combination with other elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
store at least one federated model in which the plurality of energy prediction models having high similarity are clustered (This limitation amounts to adding insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g). The step of “storing” a model is merely a generic data operation that amounts to storing and retrieving information in memory, which has been recognized by the courts as well-understood, routine, and conventional activity. See MPEP 2106.05(d)(II)(iv).).
Regarding claim 14, the rejection of claim 13 is incorporated herein. Further, claim 14 recites the following abstract ideas:
based on there being a plurality of stored federated models, separate and store the federated models for each category of a prediction result provided by the federated model (This is an abstract idea of a “mental process.” Separating models into categories requires observation and judgement. A person could review multiple records of prediction results, decide which category each belongs to, and group them accordingly. Such a classification can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas. Additionally, storing the separated groups into memory amounts to storing and retrieving information in memory, which is well-understood, routine, and conventional computer function, and further represents insignificant extra-solution activity. See MPEP 2106.05(d)(II)(iv) and 2106.05(g).).
Regarding claim 15, the following claim elements are abstract ideas:
determine whether a federated model exists in a memory of an artificial intelligence device for determining whether each of the energy prediction models is similar to the federated model (This is an abstract idea of a “mental process.” Determining whether a model exists requires observation and judgement. A model be represented as a record of household energy features, such as size of the house, appliance types, or usage history. A person could review records for a given area, observe whether another record with similar features exists, and judge the similarity between the two records. Such observation and judgement can be performed mentally or with pen and paper, and thus falls within the mental process grouping of abstract ideas. See MPEP 2106.04(a)(2)(III).);
based on determining that the federated model exists in memory, for each of the energy prediction model (This is an abstract idea of a mental process. The limitation recites a determination that can be performed in the human mind by evaluating the presence or absence of information.):
determine whether the energy prediction model is similar to the federated model by comparing gradients of the energy prediction model and gradients of the federated model (This is an abstract idea of a “mental process.” Determining similarity between models requires observation and judgement. Each model may be represented as a record of household energy features, such as a size of the house, appliance types, or usage history. After determining that a record with such features exists in a given area, a person could review the household’s own record of features alongside that record and, based on observation and judgement, determine the degree of similarity between the two. Such observation and judgement can be performed in the human mind, or with pen and paper and thus falls within the mental process grouping of abstract ideas.);
cluster the energy prediction models that are determined to be similar into the federated model to update the federated model, in order to improve the accuracy of the modeling to predict energy usage patterns (This is an abstract idea of a “mental process.” Clustering models requires observation and judgement. Each model may be represented as a record of household energy features, such as size of the house, appliance types, or usage history. A person could review a household’s record, decide whether it belongs to a particular group of similar records. This type of classification – assigning records to a group according to observed similarity – is a cognitive task that can be performed mentally or with pen and paper, and thus falls within the mental process group of abstract ideas.)
The following claim elements are additional elements which, taken alone or in combination with the other elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
receiving a plurality of energy prediction models, each of the energy prediction models before for a respective household of a plurality of households (This limitation amounts to adding insignificant extra-solution activity to the judicial exception. Receiving predictions models (i.e., mere data gathering in conjunction with the abstract idea) is directed to a well understood, routine, and conventional activity of data transmission see MPEP 2106.05(d)(II)(i).)
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.
Claims 1-4, 6, 7, 9 and 13- 15 are rejected under the 35 U.S.C. 103 as being unpatentable over Khan et al., (NPL: “An ensemble Energy Consumption Forecasting Model Based on Spatial-Temporal Clustering Analysis in Residential Buildings” (Published (2021)). in view of Obrenović et al., (NPL: “The Choice of Metric for Clustering of Electrical Power Distribution Consumers” (Published: 2017)).
Regarding claim 1, Khan discloses:
An artificial intelligence apparatus comprising for improving accuracy of modeling to predict energy usage patterns, the artificial intelligence apparatus comprising: a memory; and a processor configured to (Khan, Section 4, table 1. “Experimental environment of the proposed spatial and temporal ensemble forecasting model.” – mentions a main memory and a processor for the environmental setup. [page 3] “These days, researchers are investing more resources in improving the forecasting accuracy by employing complex and highly variable electric consumption patterns through the use of advanced ML techniques [17].”):
receive a plurality of energy prediction models, each of the energy prediction models being for a respective household of a plurality of households (Khan, [Introduction, page 2] “Recently smart metering infrastructure has provided access to high-resolution electric consumption data that has enabled researchers to analyze consumers’ load profiles using cluster analysis. Electric consumption profiles are a crucial element of energy system models.” [page 6] “Load profiles have been clustered based on similarity of consumption patterns, and models are trained considering electric load dynamic characteristics.” – Khan discloses that smart metering infrastructure provides high-resolution electric consumption data from consumers, which is processed into electric consumption (load) profiles. Consumers constitute households under the broadest reasonable interpretation. Khan further discloses that such electric consumption profiles are crucial elements of energy system models and that prediction models are trained using clustered load profiles based on similarity of consumption patterns. Accordingly, Khan teaches generating and utilizing a plurality of trained energy prediction models corresponding to household-level energy usage behavior.);
determine whether a federated model exists in the memory for determining whether each of the energy prediction models is similar to the federated model (Khan, Abstract “this paper presents a spatial and temporal ensemble forecasting model for short-term electric consumption forecasting. The proposed work involves exploring electric consumption profiles at the apartment level through cluster analysis based on the k-means algorithm. The ensemble forecasting model consists of two deep learning models; Long Short-Term Memory Unit (LSTM) and Gated Recurrent Unit (GRU). First, the apartment-level historical electric consumption data is clustered. Later the clusters are aggregated based on consumption profiles of consumers. At the building and floor level, the ensemble models are trained using aggregated electric consumption data. The proposed ensemble model forecasts the electric consumption at three spatial scales apartment, building, and floor level for hourly, daily, and weekly forecasting horizon.” – Khan discloses utilizing pre-existing prediction models at apartment, floor, and building levels and performing similarity based clustering of electric consumption profiles using distance metrics such as k-means. Such prediction models function as shared or global models derived from multiple consumption profiles. In order to perform similarity-based clustering and assignment, the system necessarily determines whether an existing shared prediction model or cluster is available for comparison prior to assigning a consumption profile. Accordingly, Khan teaches determining whether a federated model – i.e., a shared/global prediction model formed from multiple energy prediction models – exists for determining whether each energy prediction model is similar to an existing model, under BRI.);
cluster the energy prediction models that are determined to be similar to the federated model into the federated model to update the federated model, in order to improve the accuracy of the modeling to predict energy usage patterns (Khan, Section 3.2 “Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation (1). The proposed clustering mechanism employs hybrid distance metrics for the calculation of object similarity. K-means algorithm works iteratively until convergence is achieved and stable clusters are produced…Afterward, we used cluster aggregation for maximum consumption, minimum consumption, and average energy consumption among all the apartments using the equation described in the architecture diagram.” Section 3.3 “The clusters are aggregated based on minimum consumption, maximum consumption, and average consumption using the following equations…Using the aforementioned equations, we find the home having maximum, minimum, and average consumption in a given interval of time... Besides, the energy provider companies will find the home having maximum consumption and cluster all the users based on their consumption. Similarly, building and floor level electric consumption is aggregated using simple aggregation using mean and average. Afterward, the data is now ready to be fed to the ensemble forecasting model to train separate models for floor level, building level, and apartment level electric consumption forecasting.” – clustering consumption profiles based on similarity and aggregating clustered data to train forecasting models corresponds to clustering energy prediction models determined to be similar into a federated (global) mode to update the federated model.).
However, Khan does not teach, but Khan in view of Obrenović teaches the following limitation:
based on determining that the federated model exists in the memory, for each of the energy prediction models: determine whether the energy prediction model is similar to the federated model by comparing gradients of the energy prediction model and gradients of the federated model (Khan, Section 3.2 “The proposed k-mean method based on hybrid distance calculation metrics works iteratively by grouping n consumption profiles. Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation…The proposed clustering mechanism employs hybrid distance metrics for the calculation of object similarity. K-means algorithm works iteratively until convergence is achieved and stable clusters are produced.” Section 3.1 “Clustering consumption profiles of consumers based on similar load patterns and grouping them helps in partitioning data set into several groups. The similarity within a group is more significant than that among groups.” Section 3.2 “The reason for choosing an unsupervised k-mean clustering approach for our proposed ensemble forecasting model has the following objectives; finding correlations between data and potential electricity consumers to be selected for specific demand and supply programs and figuring out undesired energy consumption behaviors (i.e., outliers). The proposed k-mean method based on hybrid distance calculation metrics works iteratively by grouping n consumption profiles.” Obrenović, [pages 7-9] teaches representing consumer data as multi-dimensional vectors and determining similarity using vector-based distance metrics such as Euclidean, Manhattan, Minkowski, cosine, and correlation measures, and further explains that these vector similarity metrics are applied within k-means clustering to group consumers based on similarity between their vector representations. Gradients generated during machine-learning model training are mathematical vectors; therefore, under the broadest reasonable interpretation, comparing gradients corresponds to comparing vector-based representations. Accordingly, using k-means similarity metrics to compare model-related vectors for clustering corresponds to determining whether each energy prediction model is similar to the federated (global) model by comparing vector gradients.)
Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, having Khan and Obrenović before them, to incorporate Schneider’s teaching that daily load profiles are represented as vectors and compared using cosine distance into Khan’s ensemble clustering framework that employs k-means, Euclidean and Manhattan distance metrics. One would have been motivated to make such a combination because Khan’s formulas are inherently vector-based, and Obrenović provides explicit vector representation and dot product-based cosine similarity that would improve clarity and accuracy in measuring similarity between load profiles and their respective clusters, while also ensuring consistency across different metrics.
Regarding claim 2, Khan in view of Obrenović teaches all the elements of claim 1, therefore is rejected for the same reasons as those presented for in claim 1. Khan in view of Obrenović further teaches:
The artificial intelligence apparatus of claim 1, wherein the processor is further configured to receive the plurality of energy prediction models from an energy storage system (ESS) of each household located within a same area (Introduction “In recent times, sustainable energy generation systems are distinguished by achieving energy-efficient systems, net-zero energy buildings, implementing effective energy management strategies… Green and NetZero energy buildings based on green communication technologies are the most substantial part of sustainable urban power systems [8]. The prime focus of these systems revolve around energy-efficient buildings… Furthermore, the addition of renewables such as solar and wind-based energy generation systems, rechargeable electric gadgets, and vehicles introduce random and uncertainty to occupant electric consumption patterns [12].” – describes green and NetZero energy buildings incorporating renewables (solar, wind) which would inherently involve household-level ESS system.).
Regarding claim 3, Khan in view of Obrenović teaches all the elements of claim 1, therefore is rejected for the same reasons as those presented for in claim 1. Khan in view of Obrenović further teaches:
The artificial intelligence apparatus of claim 1, wherein the processor is further configured to, based on determining that the federated model does not exist in the memory, generate a new federated model to be stored in the memory based on an input energy prediction model (Section 3.2 “Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation (1)…K-means algorithm works iteratively until convergence is achieved and stable clusters are produced… K-mean algorithm efficiently optimize the dissimilarity with in a cluster. Using Hybrid distance metrics, the distance between the electric consumption profile and cluster centroids is computed. The purpose of using a hybrid distance metric is attributed to efficiently and objectively compute a similarity function…K-means clustering takes two input parameters; first, the number of instances to be clustered and the number of clusters to be formed.” Section 3.3 “Afterward, the data is now ready to be fed to the ensemble forecasting model to train separate models for floor level, building level, and apartment level electric consumption forecasting.” – if a household consumption profile does not match an existing cluster, the clustering process produces a new stable cluster. It is implied that these clusters are used to form or support separate ensemble models within the broader ensemble forecasting framework, which are then used for analysis. This corresponds to generating a new federated model based on an input energy prediction model when none exists.).
Regarding claim 4, Khan in view of Obrenović teaches all the elements of claim 1, therefore is rejected for the same reasons as those presented for in claim 1. Khan in view of Obrenović further teaches:
The artificial intelligence apparatus of claim 1, wherein the processor is further configured to: based on determining that the federated model exists in the memory, determine whether one or more federated models are stored in the memory (Khan, [Introduction] “We trained separate models to forecast electric consumption based on aggregated consumption data at the floor and building level. LSTM is trained using data clusters generated by the k-means algorithm for apartment level. The floor and building-level aggregated consumption data are used for model training…Moreover, we build separate models at various spatial scales, like apartment level, floor level, and building level.” – describes storing and using multiple prediction models generated at different aggregation levels (apartment, floor, and building levels). Under the broadest reasonable interpretation, maintaining multiple prediction models and selecting or applying models based on aggregation levels corresponds to determining whether one or more federated models are stored in memory after determining that a federated (global) model exists.);
based on determining that there is one federated model stored in the memory, compare an input energy prediction model and the one federated model (Khan, [Section 3.2] “Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation (1). The proposed clustering mechanism employs hybrid distance metrics for the calculation of object similarity. K-means algorithm works iteratively until convergence is achieved and stable clusters are produced… K-mean algorithm efficiently optimize the dissimilarity with in a cluster. Using Hybrid distance metrics, the distance between the electric consumption profile and cluster centroids is computed. K-means clustering takes two input parameters; first, the number of instances to be clustered and the number of clusters to be formed. We employed hybrid distance metrics for finding the distance between input data samples and the nearest centroid. The algorithm outputs cluster centroid and cluster membership for data samples based on electric consumption profiles.” – describes comparing an input consumption profile to an existing cluster centroid using similarity/distance metrics. Under BRI, comparing an input model representation to an existing cluster centroid or stored model corresponds to comparing an input energy prediction model to one federated (global) model when a single federated model is stored.).
Regarding claim 6, Khan in view of Obrenović teaches all the elements of claim 4, therefore is rejected for the same reasons as those presented in claim 4. Khan in view of Obrenović further teaches:
wherein the processor is further configured to, based on determining that the input energy prediction model and the one federated model are different from each other, generate a new federated model to be stored in the memory, based on the input energy prediction model (Khan, [Section 3.2] “Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation (1)…K-means algorithm works iteratively until convergence is achieved and stable clusters are produced… K-mean algorithm efficiently optimize the dissimilarity with in a cluster. Using Hybrid distance metrics, the distance between the electric consumption profile and cluster centroids is computed. The purpose of using a hybrid distance metric is attributed to efficiently and objectively compute a similarity function…K-means clustering takes two input parameters; first, the number of instances to be clustered and the number of clusters to be formed.” Section 3.3 “Afterward, the data is now ready to be fed to the ensemble forecasting model to train separate models for floor level, building level, and apartment level electric consumption forecasting.” – Khan teaches that if an input consumption profile does not align with an existing cluster (i.e., it is different), the k-means process inherently creates a new cluster, and that cluster is then used to train a new ensemble model.).
Regarding claim 7, Khan in view of Obrenović teaches all the elements of claim 4, therefore is rejected for the same reasons as those presented in claim 4. Khan in view of Obrenović further teaches:
The artificial intelligence apparatus of claim 4, wherein the processor is further configured to: based on determining that a plurality of federated models are stored in memory, compare the input energy prediction model and each of the plurality of federated models ; and select a federated model of the plurality of federated models having a highest similarity with respect to the input energy prediction model (Khan, section 3.2, “The reason for choosing an unsupervised k-mean clustering…finding correlations between data and potential electricity consumers to be selected for specific demand and supply programs and figuring out undesired energy consumption behaviors (i.e., outliers)...The proposed k-mean method based on hybrid distance calculation metrics works iteratively by grouping n consumption profiles…The proposed clustering mechanism employs hybrid distance metrics for the calculation of object similarity. K-means algorithm works iteratively until convergence is achieved and stable clusters are produced…The above equation merges two distance functions through average; using Manhattan distance, we can compute absolute difference between coordinate pairs while Euclidean distance finds straight line distance between two points…K-means clustering takes two input parameters; first, the number of instances to be clustered and the number of clusters to be formed. We employed hybrid distance metrics for finding the distance between input data samples and the nearest centroid. The algorithm outputs cluster centroid and cluster membership for data samples based on electric consumption profiles. The data acquired by cluster aggregation for apartment level and simple aggregation for floor and building level is used to train the ensemble forecasting model.” – describes comparing an input consumption profile to multiple cluster centroids using similarity/distance metrics and assigning the input profile to the nearest centroid. Under the broadest reasonable interpretation, comparing an input model representation to multiple stored centroids corresponds to comparing an input energy prediction model to a plurality of federated (global) models, and selecting the nearest centroid corresponds to selecting the federated model having the highest similarity.)
Regarding claim 9, Khan in view of Obrenović teaches all the elements of claim 7, therefore is rejected for the same reasons as those presented in claim 8. Khan further teaches:
wherein the processor is further configured to, based on determining that the input energy prediction model and all of the plurality of federated models are different from each other, generate a new federated model to be stored in the memory, based on the input energy prediction model (Section 3.2 “Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation (1)…K-means algorithm works iteratively until convergence is achieved and stable clusters are produced… K-mean algorithm efficiently optimize the dissimilarity with in a cluster. Using Hybrid distance metrics, the distance between the electric consumption profile and cluster centroids is computed. The purpose of using a hybrid distance metric is attributed to efficiently and objectively compute a similarity function…K-means clustering takes two input parameters; first, the number of instances to be clustered and the number of clusters to be formed.” Section 3.3 “Afterward, the data is now ready to be fed to the ensemble forecasting model to train separate models for floor level, building level, and apartment level electric consumption forecasting.” – Khan teaches that if an input consumption profile does not align with an existing cluster (i.e., it is different), the k-means process inherently creates a new cluster, and that cluster is then used to train a new ensemble model.).
Regarding claim 13, Khan in view of Obrenović teaches all the elements of claim 1, therefore is rejected for the same reasons as those presented in claim 1. Khan in view of Obrenović further teaches:
wherein the memory (Section 4, table 1. “Experimental environment of the proposed spatial and temporal ensemble forecasting model.” – mentions a main memory for the environmental setup.) is configured to store at least one federated model in which the plurality of energy prediction models having high similarity are clustered (Khan, section 3.2, “The reason for choosing an unsupervised k-mean clustering…finding correlations between data and potential electricity consumers to be selected for specific demand and supply programs and figuring out undesired energy consumption behaviors (i.e., outliers)...The proposed k-mean method based on hybrid distance calculation metrics works iteratively by grouping n consumption profiles…The proposed clustering mechanism employs hybrid distance metrics for the calculation of object similarity. K-means algorithm works iteratively until convergence is achieved and stable clusters are produced…The above equation merges two distance functions through average; using Manhattan distance, we can compute absolute difference between coordinate pairs while Euclidean distance finds straight line distance between two points…K-means clustering takes two input parameters; first, the number of instances to be clustered and the number of clusters to be formed. We employed hybrid distance metrics for finding the distance between input data samples and the nearest centroid. The algorithm outputs cluster centroid and cluster membership for data samples based on electric consumption profiles. The data acquired by cluster aggregation for apartment level and simple aggregation for floor and building level is used to train the ensemble forecasting model.” – first determine similarity between input profile and multiple clusters and then using distance metrics to measure the similarity in that cluster (highest similarity). Training the model implies updating a model which inherently requires storing the model, since the very purpose of training a model is to generate and retain the updated version for continued use.)
Regarding claim 14, Khan in view of Obrenović teaches all the elements of claim 13, therefore is rejected for the same reasons as those presented in claim 13. Khan in view of Obrenović further teaches:
wherein the memory (Section 4, table 1. “Experimental environment of the proposed spatial and temporal ensemble forecasting model.” – mentions a main memory for the environmental setup.) is further configured to, based on there being a plurality of stored federated models, separate and store the federated models for each category of a prediction result provided by the federated model (Khan, page 4, Introduction “Our proposed spatial and temporal ensemble forecasting model integrates an unsupervised k-mean clustering approach with a supervised learning approach to produce accurate consumption forecasts at the apartment level. We trained separate models to forecast electric consumption based on aggregated consumption data at the floor and building level. LSTM is trained using data clusters generated by the k-means algorithm for apartment level. The floor and building-level aggregated consumption data are used for model training…Moreover, we build separate models at various spatial scales, like apartment level, floor level, and building level. We produced electric consumption forecasting at three significant temporal scales including hourly, daily and weekly forecasting horizon. All the stated reasons established our choice of learners and proper network architecture selection for enhancing the accuracy of our proposed models.” – multiple models are created and maintained separately according to categories of prediction results, such as apartment-level, floor-level, and building level consumption forecasts, with each category having its own stored model.)
Regarding claim 15, Khan discloses:
A method for improving accuracy of modeling to predict energy usage patterns (Khan, [page 3, Introduction] “These days, researchers are investing more resources in improving the forecasting accuracy by employing complex and highly variable electric consumption patterns through the use of advanced ML techniques [17].”):, the method comprising:
receiving a plurality of energy prediction models, each of the energy prediction models being for a respective household of a plurality of households (Khan, [Introduction, page 2] “Recently smart metering infrastructure has provided access to high-resolution electric consumption data that has enabled researchers to analyze consumers’ load profiles using cluster analysis. Electric consumption profiles are a crucial element of energy system models.” [page 6] “Load profiles have been clustered based on similarity of consumption patterns, and models are trained considering electric load dynamic characteristics.” – Khan discloses that smart metering infrastructure provides high-resolution electric consumption data from consumers, which is processed into electric consumption (load) profiles. Consumers constitute households under the broadest reasonable interpretation. Khan further discloses that such electric consumption profiles are crucial elements of energy system models and that prediction models are trained using clustered load profiles based on similarity of consumption patterns. Accordingly, Khan teaches generating and utilizing a plurality of trained energy prediction models corresponding to household-level energy usage behavior.);
determining whether a federated model exists in a memory of an artificial intelligence device for determining whether each of the energy prediction models is similar to the federated model (Abstract “this paper presents a spatial and temporal ensemble forecasting model for short-term electric consumption forecasting. The proposed work involves exploring electric consumption profiles at the apartment level through cluster analysis based on the k-means algorithm. The ensemble forecasting model consists of two deep learning models; Long Short-Term Memory Unit (LSTM) and Gated Recurrent Unit (GRU). First, the apartment-level historical electric consumption data is clustered. Later the clusters are aggregated based on consumption profiles of consumers. At the building and floor level, the ensemble models are trained using aggregated electric consumption data. The proposed ensemble model forecasts the electric consumption at three spatial scales apartment, building, and floor level for hourly, daily, and weekly forecasting horizon.” – describes storing and using cluster-based and spatial-scale forecasting models generated from clustered consumption profiles. The clustering process relies on previously generated cluster centroids and stored model representations to evaluate the similarity between input consumption profiles and existing clustered models. Under BRI, using stored cluster centroids or aggregated forecasting models for similarity comparison corresponds to determining whether a federated model exists in memory for determining similarity between energy prediction models.);
clustering the energy prediction models that are determined to be similar to the federated model into the federated model to update the federated model, in order to improve the accuracy of the modeling to predict energy usage patterns (Section 3.2 “Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation (1). The proposed clustering mechanism employs hybrid distance metrics for the calculation of object similarity. K-means algorithm works iteratively until convergence is achieved and stable clusters are produced…Afterward, we used cluster aggregation for maximum consumption, minimum consumption, and average energy consumption among all the apartments using the equation described in the architecture diagram.” Section 3.3 “The clusters are aggregated based on minimum consumption, maximum consumption, and average consumption using the following equations…Using the aforementioned equations, we find the home having maximum, minimum, and average consumption in a given interval of time... Besides, the energy provider companies will find the home having maximum consumption and cluster all the users based on their consumption. Similarly, building and floor level electric consumption is aggregated using simple aggregation using mean and average. Afterward, the data is now ready to be fed to the ensemble forecasting model to train separate models for floor level, building level, and apartment level electric consumption forecasting.” – clustering consumption profiles based on similarity and aggregating clustered data to train forecasting models corresponds to clustering energy prediction models determined to be similar into a federated (global) mode to update the federated model.).
However, Khan does not teach but Khan in view of Obrenović teaches the following limitation:
based on determining that the federated model exists in the memory, for each of the energy prediction models: determining whether the energy prediction model is similar to the federated model by comparing gradients of the energy prediction model and the gradients of the federated model (Khan, Section 3.2 “The proposed k-mean method based on hybrid distance calculation metrics works iteratively by grouping n consumption profiles. Each consumption profile represents hourly consumption data acquired through smart meters grouped into k clusters by minimizing the sum of squared among clusters demonstrated in Equation…The proposed clustering mechanism employs hybrid distance metrics for the calculation of object similarity. K-means algorithm works iteratively until convergence is achieved and stable clusters are produced.” Section 3.1 “Clustering consumption profiles of consumers based on similar load patterns and grouping them helps in partitioning data set into several groups. The similarity within a group is more significant than that among groups.” Section 3.2 “The reason for choosing an unsupervised k-mean clustering approach for our proposed ensemble forecasting model has the following objectives; finding correlations between data and potential electricity consumers to be selected for specific demand and supply programs and figuring out undesired energy consumption behaviors (i.e., outliers). The proposed k-mean method based on hybrid distance calculation metrics works iteratively by grouping n consumption profiles.” Obrenović, [pages 7-9] teaches representing consumer data as multi-dimensional vectors and determining similarity using vector-based distance metrics such as Euclidean, Manhattan, Minkowski, cosine, and correlation measures, and further explains that these vector similarity metrics are applied within k-means clustering to group consumers based on similarity between their vector representations. Gradients generated during machine-learning model training are mathematical vectors; therefore, under the broadest reasonable interpretation, comparing gradients corresponds to comparing vector-based representations. Accordingly, using k-means similarity metrics to compare model-related vectors for clustering corresponds to determining whether each energy prediction model is similar to the federated (global) model by comparing vector gradients.)
Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, having Khan and Obrenović before them, to incorporate Schneider’s teaching that daily load profiles are represented as vectors and compared using cosine distance into Khan’s ensemble clustering framework that employs k-means, Euclidean and Manhattan distance metrics. One would have been motivated to make such a combination because Khan’s formulas are inherently vector-based, and Obrenović provides explicit vector representation and dot product-based cosine similarity that would improve clarity and accuracy in measuring similarity between load profiles and their respective clusters, while also ensuring consistency across different metrics.
Claims 12 is rejected under the 35 U.S.C. 103 as being unpatentable over Khan et al., (NPL: “An ensemble Energy Consumption Forecasting Model Based on Spatial-Temporal Clustering Analysis in Residential Buildings” (Published (2021)). in view of Obrenović et al., (NPL: “The Choice of Metric for Clustering of Electrical Power Distribution Consumers” (Published: 2017)) further in view of Liu et al., (NPL: “Converse-Averse Gradient Descent for Mult-task Learning” (Published: 2021)).
Regarding claim 12, Khan in view of Obrenović teaches all the elements of claim 1, therefore is rejected for the same reasons as those presented in claim 10. Khan in view of Obrenović does not teach, but Liu does teach:
wherein the processor is further configured to update the federated model based on a project conflicting gradients (PCGrad) algorithm (Introduction “Motivated by the limitation of current methods, we introduce Conflict-Averse Gradient descent (CAGrad), which reduces the conflict among gradients and still provably converges to a minimum of the average loss. The idea of CAGrad is simple: it looks for an update vector that maximizes the worst local improvement of any objective in a neighborhood of the average gradient. In this way, CAGrad automatically balances different objectives and smoothly converges to an optimal point of the average loss.”)
Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date, having Khan and Liu before them, to apply a gradient-conflict resolution algorithm such as CAGrad when updating ensemble models. One would have been motivated to make such a modification in order to reduce conflicting gradients and balance multiple objectives during training, thereby improving stability and convergence when updating the ensemble model.
Conclusion
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 Daravanh Phakousonh whose telephone number is (571)272-6324. The examiner can normally be reached Mon - Thurs 7 AM - 5 PM, Every other Friday 7 AM - 4PM.
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, Li B Zhen can be reached at 571-272-3768. 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.
/Daravanh Phakousonh/Examiner, Art Unit 2121
/Li B. Zhen/Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights