Systems and methods for processing data relating to energy usage

A system comprising:

• a server;
• and
• a memory storing instructions configured to instruct the server to perform operations comprising:
• automatically importing different types or forms of data from a plurality of different data sources, wherein at least a portion of the imported data is associated with at least one of a sensor and a meter belonging to said plurality of different data sources;
• converting the imported data based on a data model, wherein the data model comprises a set of canonical object definitions, wherein the canonical object definitions include standardized interface definitions based on industry standards;
• normalizing the converted data by filling in gaps in the converted data and addressing outliers in the converted data;
• persisting the normalized data in a plurality of different data stores comprising a relational database and a key-value store;
• and
• performing analytics on the persisted data, wherein the performing of said analytics comprises:
• (i) a prediction step comprising feature extraction, classification, and ranking, wherein said feature extraction comprises providing the persisted data to a feature extraction machine learning model comprising one or more learning detection parameters, wherein the feature extraction machine learning model is configured to (a) extract a plurality of features associated with signatures of malfunction and/or theft from the persisted data and (b) identify one or more data sources that exhibit the signatures of malfunction and/or theft, wherein said classification comprises providing the one or more data sources identified as having the signatures of malfunction or theft to a classification model configured to selectively merge two or more related features of the plurality of features or apply different weights to the plurality of features based on signature types in order to classify each of the one or more data sources as having a signature of malfunction or a signature of theft, and wherein said ranking comprises generating a priority order for resolving the malfunction and/or theft of the one or more data sources based on an impact of the malfunction and/or theft on business operations of an entity, and
• (ii) using empirical data as supervised feedback to adjust the one or more learning detection parameters and train the feature extraction machine learning model, wherein the empirical data comprises data generated from investigation of the plurality of different data sources by field personnel.

A non-transitory computer-readable storage medium including instructions that, when executed by at least one processor of a computing system, cause the computing system to perform operations comprising:

• automatically importing different types or forms of data from a plurality of different data sources, wherein at least a portion of the imported data is associated with at least one of a sensor and a meter belonging to said plurality of different data sources;
• converting the imported data based on a data model, wherein the data model comprises a set of canonical object definitions, wherein the canonical object definitions include standardized interface definitions based on industry standards;
• normalizing the converted data by filling in gaps in the converted data and addressing outliers in the converted data;
• persisting the normalized data in a plurality of different data stores comprising a relational database and a key-value store;
• and
• performing analytics on the persisted data, wherein the performing of said analytics comprises:
• (i) a prediction step comprising feature extraction, classification, and ranking, wherein said feature extraction comprises providing the persisted data to a feature extraction machine learning model comprising one or more learning detection parameters, wherein the feature extraction machine learning model is configured to (a) extract a plurality of features associated with signatures of malfunction and/or theft from the persisted data and (b) identify one or more data sources that exhibit the signatures of malfunction and/or theft, wherein said classification comprises providing the one or more data sources identified as having the signatures of malfunction or theft to a classification model configured to selectively merge two or more related features of the plurality of features or apply different weights to the plurality of features based on signature types in order to classify each of the one or more data sources as having a signature of malfunction or a signature of theft, and wherein said ranking comprises generating a priority order for resolving the malfunction and/or theft of the one or more data sources based on an impact of the malfunction and/or theft on business operations of an entity, and
• (ii) using empirical data as supervised feedback to adjust the one or more learning detection parameters and train the feature extraction machine learning model, wherein the empirical data comprises data generated from investigation of the plurality of different data sources by field personnel.