Composite event estimation through temporal logic
View Patent ↗A computer-implemented method of discovering a composite durational event structure through temporal logic includes identifying a plurality of temporally related atomic events from temporal data trajectories of a multivariate dataset according to a definition of an atomic event predicate. At least one composite event having a durational event structure of at least some of the plurality of the temporally related atomic events is discovered by machine learning. An action is performed that is selected from a predetermined list associated with the composite event.
1. A computer-implemented method of discovering a composite durational event structure through temporal logic to predict, mitigate, or prevent a failure condition of an electrical grid system, the method comprising:
for each data modality of a time series, transforming the time series data into a multivariate dataset that can be automatically queried, using one or more modality-specific preprocessors, wherein each modality-specific preprocessor applies a structured transformation based on predefined rules tailored to the respective modality;
identifying, via a trained machine learning model, a plurality of temporally related atomic events from temporal data trajectories of the multivariate dataset, according to a definition of an atomic event predicate rule that specifies temporal relationships and threshold conditions;
discovering, by the machine learning model, at least one composite event having a durational event structure of at least some of the plurality of the temporally related atomic events, wherein the durational event structure is determined based on machine learned time-dependent correlations between atomic events;
generating, by the trained machine learning model, a composite event classification by combining predicate rules that determine which of the plurality of temporally related atomic events make up the at least one composite event, wherein the predicate rules are learned and refined by the trained machine learning model based on historical event patterns;
performing a system control action of the electrical grid system selected from a predetermined list associated with the at least one composite event comprising at least one of mitigating a current failure of the electrical grid system or preventing an impending failure of the electrical grid system, based on the composite event classification and the identified atomic events that make up the composite event, wherein:
the at least one composite event comprises a failure of the electrical grid system; and
the temporally related atomic events include real-time sensor data collected from distributed components of the one or more components of the electrical grid system.
2. The computer-implemented method of claim 1 , wherein the performed action comprises notifying a designated entity from the list associated with the composite event.
3. The computer-implemented method of claim 1 , wherein the temporal data trajectories of the multivariate dataset from which the atomic events are identified by determining input trajectories comprise at least two variables measured numerically across a time interval.
4. The computer implemented method of claim 1 , wherein the atomic event predicate is determined based on a domain-specific knowledge.
5. The computer-implemented method of claim 4 , wherein the multivariate data set comprises a video dataset, and the identifying of the temporally related atomic events further comprises:
processing a raw timeseries data of the multivariate dataset; and
identifying individual atomic events having a duration of a sub-interval of a time interval of the raw time series data.
6. The computer-implemented method of claim 5 , further comprising additionally determining the atomic event predicate by processing the raw time series data of the multivariate dataset.
7. The computer-implemented method of claim 4 ,
wherein the multivariate dataset comprises raw data; and
the method further comprising learning a corresponding rule structure based on a labeling of the raw data.
8. The computer-implemented method of claim 7 , further comprising:
ingesting temporal trajectories in the raw data of the multivariate dataset; and
storing the ingested temporal trajectories as time trajectories in an automatically chosen state space construction.
9. The computer-implemented method of claim 4 , further comprising identifying the at least one composite event by constructing a timeline of temporally related atomic events, along with their temporal relation.
10. The computer-implemented method of claim 9 , further comprising:
annotating the timeline based on the atomic event predicate; and
localizing one or more sub-intervals on the timeline corresponding to the atomic events.
11. The computer-implemented method of claim 10 , further comprising learning the durational event structure of the at least one composite event using the localized sub-intervals on the timeline in a supervised learning operation.
12. The computer-implemented method of claim 10 , further comprising learning the durational event structure of the at least one composite event using the localized sub-intervals on the timeline in a reinforcement learning operation.
13. The computer-implemented method of claim 1 , further comprising:
applying the discovered durational event structure of the at least one composite event by analyzing the real-time streaming data; and
detecting an evolving progression of a particular composite event by detecting constituent atomic events and verifying the temporal relations among the atomic events.
14. A computing device configured to discover a composite durational event structure through temporal logic to predict, mitigate, or prevent a failure condition of an electrical grid system, the computing device comprising:
a processor;
a memory coupled to the processor, the memory storing instructions to cause the processor to perform acts comprising:
for each data modality of a time series, transforming the time series data into a multivariate dataset that can be automatically queried, using one or more modality-specific preprocessors, wherein each modality-specific preprocessor applies a structured transformation based on predefined rules tailored to the respective modality;
identifying, via a trained machine learning model, a plurality of temporally related atomic events from temporal data trajectories of the multivariate dataset according to a definition of an atomic event predicate rule that specifies temporal relationships and threshold conditions;
discovering, by the machine learning model, at least one composite event having a durational event structure of at least some of the plurality of the temporally related atomic events, wherein the durational event structure is determined based on machine learned time-dependent correlations between atomic events;
generating, by the trained machine learning model, a composite event classification by combining predicate rules that determine which of the plurality of temporally related atomic events make up the at least one composite event, wherein the predicate rules are learned and refined by the trained machine learning model based on historical event patterns; and
performing an action selected from a predetermined list associated with the composite event comprising at least one of mitigating a current failure of the electrical grid system or preventing a failure of the electrical grid system, wherein:
the at least one composite event comprises a failure of the electrical grid system; and
the temporally related atomic events include real-time sensor data collected from distributed components of the one or more components of the electrical grid system.
15. The computing device of claim 14 , wherein the instructions cause the processor to perform additional acts, comprising:
identifying the at least one composite event by constructing a timeline of temporally related atomic events, along with the temporal relations among the atomic events;
annotating the timeline using the atomic event predicate; and
localizing sub-intervals on the timeline corresponding to the temporally related atomic events.
16. The computing device of claim 14 ,
wherein the atomic event predicate is based on a domain-specific knowledge; and
the instructions cause the processor to perform additional acts comprising:
identifying the temporally related atomic events by processing a raw timeseries data of the multivariate dataset; and
identifying individual atomic events having a duration of a sub-interval of a time interval of the raw time series data.
17. The computing device of claim 16 , wherein the multivariate dataset includes at least one of a video data set and an audio data set, and the instructions cause the processor to perform an additional act comprising:
learning the durational event structure of the at least one composite event using the localized sub-intervals on the time line in a reinforcement learning operation.
18. The computing device of claim 16 , wherein the instructions cause the processor to perform an additional act comprising:
learning the durational event structure of the at least one composite event using the localized sub-intervals on the timeline in a supervised learning operation.
19. A non-transitory computer readable storage medium tangibly embodying a computer readable program code having computer readable instructions that, when executed, causes a computer device to carry out a method of discovering a composite durational event structure through temporal logic to predict, mitigate, or prevent a failure condition of an electrical grid system, the method comprising:
for each data modality of a time series, transforming the time series data into a multivariate dataset that can be automatically queried, using one or more modality-specific preprocessors, wherein each modality-specific preprocessor applies a structured transformation based on predefined rules tailored to the respective modality;
identifying, via a trained machine learning model, a plurality of temporally related atomic events from temporal data trajectories of the multivariate dataset, according to a definition of an atomic event predicate rule that specifies temporal relationships and threshold conditions;
discovering, by the machine learning model, at least one composite event having a durational event structure of at least some of the plurality of the temporally related atomic events, wherein the durational event structure is determined based on machine learned time-dependent correlations between atomic events;
generating, by the trained machine learning model, a composite event classification by combining predicate rules that determine which of the plurality of temporally related atomic events make up the at least one composite event, wherein the predicate rules are learned and refined by the trained machine learning model based on historical event patterns; and
performing an action selected from a predetermined list associated with the composite event comprising at least one of mitigating a current failure of an electrical grid system or preventing a failure of the electrical grid system.