Anomaly Detection via Normal Behavior Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems rely on training data from known-anomalous behavior, limiting their deployment speed and applicability in systems that fail rarely, where little information is available about anomalous behavior.
Innovation Solution
A method that predicts normal behavior of a monitored system using only sensor data collected during normal behavior, allowing for the detection and correction of abnormal behavior by comparing predicted behavior to recent sensor data and performing corrective actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If anomaly detection systems use training data from known-anomalous behavior, then detection accuracy for known anomalies is improved, but deployment speed is reduced and applicability to rare failures is limited
Solution Approach 1:
Instead of training the model to detect anomalies directly using anomalous data, the patent inverts the approach by training the model to predict normal behavior using only normal operation data. The anomaly detection then emerges from deviations between predicted normal behavior and actual system behavior, eliminating the need for scarce anomalous training data and enabling faster deployment.
Solution Approach 2:
The patent introduces a normal behavior prediction model as an intermediary between the system being monitored and the anomaly detection process. This mediator predicts what normal behavior should look like, and anomalies are detected as deviations from this prediction, rather than directly comparing against known anomaly patterns.
2Measurement precision
If anomaly detection systems use training data from known-anomalous behavior, then detection accuracy for known anomalies is improved, but applicability to systems with rare failures is reduced
Solution Approach 1:
The patent inverts the traditional anomaly detection paradigm by training exclusively on normal behavior data rather than anomalous data. This inversion enables the system to detect any deviation from normal behavior, including rare and previously unseen anomalies, making it highly adaptable to systems where failure data is scarce or nonexistent.
Solution Approach 2:
The normal behavior prediction model serves multiple functions: it detects known anomalies, identifies rare failures, and generalizes to previously unseen anomaly types. By learning the underlying patterns of normal operation, the model becomes universally applicable across different failure modes without requiring retraining on specific anomaly types.
Data Source
AI summary
Methods and systems for detecting and correcting anomalies include predicting normal behavior of a monitored system based on training data that includes only sensor data collected during normal behavior of the monitored system. The predicted normal behavior is compared to recent sensor data to determine that the monitored system is behaving abnormally. A corrective action is performed responsive to the abnormal behavior to correct the abnormal behavior.