Methods and apparatus for detection of process parameter anomalies
AssigneeArizona Public Service Company
InventorsJerrold Vincent, Bradley Fox
Life of this patent
Estimating the term…
Abstract
Methods and apparatus for identifying anomalies in data may operate in conjunction with a processing system and an output. The system may receive actual data and generate nominal data according to the actual data. The system may compare the actual data to the nominal data and identify an outlier in the actual data compared to the nominal data. The output may provide information relating to the identified outlier. In various embodiments, the system may include a neural network.
Claims33
A system for identifying anomalies in actual time series data from multiple sources, comprising:
- a processing system, implemented via one or more computer systems, configured to receive the actual time series data, wherein the processing system comprises a neural network generating a nominal predictive time series dataset according to the actual time series data, and wherein the processing system:
- compares the actual time series data to the nominal predictive time series dataset;
- and identifies an outlier in the actual time series data compared to the nominal predictive time series dataset;
- and an output responsive to the processing system and providing information relating to the identified outlier.
A system for identifying anomalies in multiple streams of actual time series data from multiple power generation sources, comprising:
- a recurrent auto-encoding neural network configured to receive the actual time series data, wherein the recurrent auto-encoding neural network generates a multivariate nominal predictive dataset based on a group of the streams of the actual time series data from the multiple power generation sources;
- a processing system, implemented via one or more computer systems, responsive to the recurrent auto-encoding neural network, wherein the processing system:
- compares the multivariate nominal predictive dataset data sot to an actual multivariate dataset based on the group of the streams of the actual time series data from the multiple power generation sources to generate a delta dataset;
- and identifies an outlier in the delta dataset;
- and an output responsive to the processing system and providing information relating to the identified outlier.
A method for identifying anomalies in multiple streams of actual time series data from multiple power generation sources, comprising:
- generating a multivariate nominal predictive dataset based on a group of the streams of the actual time series data from multiple power generation sources;
- generating a delta dataset according to a comparison of the multivariate nominal predictive dataset to an actual multivariate dataset based on the group of the streams of the actual time series data from the multiple power generation sources;
- identifying an outlier in the delta dataset;
- and
- providing information relating to the identified outlier at an output.
Dependencies are inferred from the claim text.