Anomaly Detection via Stable Training Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex physical systems with thousands of interconnected components generate noisy and contradictory time series data, making it challenging to effectively analyze and identify anomalies, which can lead to system failures.
Innovation Solution
A computer-implemented anomaly detection method that receives sensor data from multiple sensors, generates a relationship model using pairs of time series, updates the model with new data, and identifies anomalies based on a fused single-variant time series fitness score, controlling machine operations in response to detected anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data from multiple sensors is collected to monitor complex physical systems, then system monitoring capability is improved, but data noise and contradiction increase
Solution Approach 1:
The patent combines multiple time series data from different sensors into a fused single-variant time series. This merging process integrates information from multiple sources while applying fusion algorithms to reduce noise and contradictions, transforming heterogeneous sensor data into a unified representation that preserves essential system state information.
Solution Approach 2:
The relationship model acts as an intermediary between raw sensor data and anomaly detection. This model learns stable relationships between different time series and serves as a mediator to filter out noise and contradictions, allowing the system to detect anomalies based on deviations from learned normal relationships rather than directly processing noisy raw data.
2Measurement precision
If relationship model is continuously updated with new sensor data, then anomaly detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The relationship model is designed to be dynamic and adaptive, continuously updating its parameters as new sensor data arrives. This dynamic updating allows the model to adapt to changing system conditions and improve anomaly detection accuracy over time, while the incremental update approach avoids the need to reprocess all historical data, managing computational complexity.
Solution Approach 2:
The system performs preliminary learning of stable relationships between time series during normal operation before anomaly detection is needed. By pre-learning these relationships when data is clean and representative, the system prepares the relationship model in advance, reducing the computational burden during actual anomaly detection phases.
3Measurement precision
If stable training regions are selected for model training, then model accuracy is improved, but training data requirements increase
Solution Approach 1:
The patent segments the time series data into different regions based on stability characteristics. Instead of using all available data uniformly, the system identifies and selects only the stable training regions where relationships between variables are consistent and reliable. This segmentation approach improves model accuracy by using high-quality training data while reducing the overall quantity of training data needed.
Solution Approach 2:
The system applies different quality standards to different portions of the training data. Stable regions are identified and given higher weight or exclusive use for training, while unstable regions are excluded or downweighted. This local quality approach ensures that the model learns from the most reliable data segments, improving accuracy without requiring uniformly high-quality data across the entire dataset.
Data Source
AI summary
A computer-implemented method, system, and computer program product are provided for anomaly detection. The method includes receiving, by a processor, sensor data from a plurality of sensors in a system. The method also includes generating, by the processor, a relationship model based on the sensor data. The method additionally includes updating, by the processor, the relationship model with new sensor data. The method further includes identifying, by the processor, an anomaly based on a fused single-variant time series fitness score in the relationship model. The method also includes controlling an operation of a processor-based machine to change a state of the processor-based machine, responsive to the anomaly.