Transformer-Based Anomaly Detection for Time Series and Event Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to accurately predict anomalies in complex systems like cyber-physical and healthcare systems, relying solely on time series or event history, which limits their effectiveness in preventative maintenance and adverse event prediction.
Innovation Solution
A method and system using transformers and an aggregation network to encode multivariate time series and multi-type event sequences, generating a feature vector for anomaly detection, followed by corrective actions to mitigate anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only time series data is used for anomaly detection, then the system is simpler to implement, but the detection accuracy is insufficient
Solution Approach 1:
The patent combines time series data processing with event sequence processing into a unified anomaly detection system. The encoder integrates both data types, merging their representations to jointly detect anomalies, thereby improving detection accuracy while managing system complexity through a cohesive architecture.
Solution Approach 2:
The encoder is designed to handle multiple data types (time series and events) universally. It processes both continuous time series measurements and discrete event sequences through the same neural network architecture, enabling the system to leverage diverse data sources for improved anomaly detection without requiring separate specialized processors.
2Measurement precision
If only event history is used for anomaly detection, then data processing is simpler, but the ability to detect gradual anomalies is limited
Solution Approach 1:
The system merges event sequence analysis with time series analysis to detect anomalies. The encoder combines representations from both data sources, enabling the detection of gradual anomalies through time series data while incorporating contextual information from event sequences, thereby improving overall detection accuracy.
Solution Approach 2:
The patent segments the anomaly detection task into two parallel processing streams: one for time series data and one for event sequences. Each stream is processed independently by the encoder, allowing specialized handling of each data type's characteristics while integrating their results for comprehensive anomaly detection.
3Reliability
If both time series and event sequences are processed separately, then each can be optimized independently, but the integration of insights is difficult
Solution Approach 1:
The encoder merges the processed representations from time series and event sequences into a unified feature space. This integration allows the system to leverage insights from both data sources simultaneously, improving detection reliability by considering multiple perspectives on system anomalies within a single coherent model.
Data Source
AI summary
Methods and systems for anomaly detection include encoding a multivariate time series and a multi-type event sequence using respective transformers and an aggregation network to generate a feature vector. Anomaly detection is performed using the feature vector to identify an anomaly within a system. A corrective action is performed responsive to the anomaly to correct or mitigate an effect of the anomaly. The detected anomaly can be used in a healthcare context to support decision making by medical professionals with respect to the treatment of a patient. The encoding may include machine learning models to implement the transformers and the aggregation network using deep learning.