Multivariate Time-Series Precursor Detection for Early Anomaly Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Early anomaly detection in multi-variate time series data is challenging due to uncertainty in identifying the exact time when precursor symptoms occur, making it difficult to use user-provided labeled data effectively for predictive maintenance.
Innovation Solution
The method employs multi-instance learning with long short-term memory (LSTM) to identify precursor feature vectors from multi-variate time series data segments before an abnormal period, using contrastive loss and hinge loss to ensure the extracted features are robust and interpretable, even with limited labeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If user-provided labeled abnormal time periods are used for anomaly detection, then the detection can be performed with available data, but the exact timing of precursor symptoms remains uncertain and difficult to identify
Solution Approach 1:
The patent segments the time series data into multiple instance sections within the abnormal time period. Each instance section is independently analyzed to identify precursor symptoms, allowing the system to pinpoint specific time segments where anomalies begin rather than treating the entire abnormal period as a single unit. This segmentation resolves the timing uncertainty by breaking down the ambiguous abnormal period into analyzable segments.
Solution Approach 2:
The patent performs preliminary analysis by extracting precursor features from instance sections that occur before the confirmed abnormal phase. By analyzing the time series data in reverse chronological order from the labeled abnormal period, the system identifies precursor symptoms in earlier time segments, enabling early detection before the full anomaly manifests. This preliminary action allows the system to detect anomalies at their earliest detectable stage.
2Measurement precision
If multi-instance learning with LSTM is applied to extract precursor features, then early detection accuracy is improved, but the computational complexity and model requirements increase
Solution Approach 1:
The patent divides the time series data into multiple instance sections and applies multi-instance learning, where each instance section is processed independently by the LSTM model. This segmentation allows the complex LSTM model to focus on smaller, manageable time segments rather than processing the entire time series at once, making the computational task more tractable while maintaining high detection accuracy through aggregated results from multiple instances.
Solution Approach 2:
The patent applies LSTM modeling selectively to extract only the relevant precursor features from each instance section, rather than attempting to model all aspects of the time series data. By focusing computational resources on identifying specific precursor patterns rather than comprehensive analysis, the system achieves high detection accuracy with reduced computational overhead compared to full-series modeling.
Data Source
AI summary
Systems and methods for early anomaly prediction on multi-variate time series data are provided. The method includes identifying a user labeled abnormal time period that includes at least one anomaly event. The method also includes determining a multi-variate time series segment of multivariate time series data that occurs before the user labeled abnormal time period, and treating, by a processor device, the multi-variate time series segment to include precursor symptoms of the at least one anomaly event. The method includes determining instance sections from the multi-variate time series segment and determining at least one precursor feature vector associated with the at least one anomaly event for at least one of the instance sections based on applying long short-term memory (LSTM). The method further includes dispatching predictive maintenance based on the at least one precursor feature vector.