Anomaly Sign Detection Model with Correlation-Based Data Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly sign detection techniques using AI in power plants face challenges with false positives due to insufficient data, leading to unreliable anomaly detection, especially when seasonal variations and operational changes occur, causing normal states to be incorrectly identified as abnormal.
Innovation Solution
An anomaly sign detection system that uses machine learning to correct process values by determining correlation with reference values, generating learning input data, and employing an autoencoder to suppress false positives and improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If unsupervised learning is applied to detect anomalies in power plant data, then the system can handle infinite abnormal states without predefined indices, but normal states under seasonal variations and operational changes are incorrectly identified as anomalies
Solution Approach 1:
The system performs preliminary data correction before anomaly detection by calculating correction values based on historical correlations between process values. This preliminary action removes seasonal variations and operational changes from the data, allowing the unsupervised learning model to focus on actual anomalies without being misled by normal variations.
Solution Approach 2:
The patent introduces correction values as an intermediary element between the raw process data and the anomaly detection model. These correction values, derived from historical correlation analysis, mediate the relationship by adjusting the data to account for seasonal and operational variations, thereby improving detection reliability.
2Quantity of substance
If data augmentation techniques such as adding noise or changing frequency components are applied to time-series data, then the amount of learning data increases, but the physical constraints of process values (certain ranges and time periods) are violated
Solution Approach 1:
Instead of applying conventional data augmentation techniques that add noise or alter frequency components, the patent changes the parameter of data representation by generating corrected process values based on historical correlations. This approach increases data quantity while maintaining physical validity, as the corrections are derived from actual operational patterns rather than random transformations.
3Quantity of substance
If conventional data augmentation methods are used to compensate for insufficient learning data, then more data becomes available for training, but false positive anomaly detections increase due to inadequate representation of normal operational variations
Solution Approach 1:
The system performs preliminary correction of process values using historical correlation data before feeding them to the anomaly detection model. This preliminary action ensures that normal operational variations due to seasonal changes and operational conditions are accounted for, reducing false positives while providing sufficient training data.
Solution Approach 2:
The patent incorporates feedback from historical operational data by continuously calculating correction values based on past correlations between process values. This feedback mechanism allows the system to learn from historical patterns and improve its ability to distinguish between normal variations and actual anomalies, thereby reducing false positives.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
According to one embodiment, an anomaly sign detection system comprising one or more computers configured to: calculate a correction value for correcting at least one actual process value from the at least one actual process value and at least one reference process value; determine whether each of plurality of actual process values is correlated with the at least one reference process value or not, based on correction-necessity coefficient of determination; use the correction value for correcting at least one actual process value determined to be correlated with the at least one reference process value among the plurality of actual process values; generate learning input data including at least one corrected process value as the at least one actual process value corrected by the correction value; and perform machine learning by inputting the learning input data to anomaly sign detection model.