Energy Storage Anomaly Detection Using Unsupervised Learning Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality detection systems for energy storage devices face challenges in preparing appropriate learning data, as it is difficult to separate normal and abnormal data, especially when the energy storage device is new or its environment changes, leading to incorrect detection of abnormalities.
Innovation Solution
An abnormality detection device that creates learning data by statistically processing measurement data, including both normal and abnormal data, using unsupervised learning methods like autoencoders, which simplifies the preparation of learning data and reduces the influence of time and environmental differences between data collection and detection periods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning methods are used to create abnormality detection models, then detection accuracy can be improved, but the complexity of preparing learning data increases significantly
Solution Approach 1:
Instead of using supervised learning that requires separating normal and abnormal data (conventional approach), the patent inverts the approach by using unsupervised learning on only normal data. The model learns normal patterns, and any deviation from these patterns is detected as abnormality, eliminating the need for abnormal data collection and labeling.
Solution Approach 2:
The patent extracts only the normal data component from the learning dataset, excluding abnormal data entirely. By focusing solely on normal operational patterns, the system simplifies data preparation while maintaining detection capability through anomaly detection of deviations from the learned normal state.
2Measurement precision
If abnormal data is excluded from learning data to improve model accuracy, then detection precision improves, but the time required for data preparation increases
Solution Approach 1:
The patent extracts and uses only normal data for training, completely excluding abnormal data from the learning process. This extraction approach reduces data preparation time by eliminating the need to identify, separate, and label abnormal data, while still achieving high detection precision through unsupervised anomaly detection.
Solution Approach 2:
The system performs self-service by automatically learning normal patterns without requiring manual classification of data into normal and abnormal categories. The unsupervised learning algorithm autonomously identifies deviations from normal operation, eliminating time-consuming manual data preparation steps.
3Reliability
If traditional abnormality detection methods are used, then detection capability is maintained, but the system cannot adapt when energy storage device characteristics change over time
Solution Approach 1:
The patent implements dynamic adaptability by periodically retraining the unsupervised learning model with newly collected normal data. This allows the model to adapt to changing device characteristics over time, maintaining detection reliability while accommodating temporal variations in operational patterns.
Solution Approach 2:
The unsupervised learning model serves multiple functions: it detects known abnormalities, adapts to characteristic changes, and generalizes to various operational conditions without requiring reconfiguration. This universal approach maintains detection capability across different device states and time periods.
Data Source
AI summary
An abnormality detection device includes: a creation unit that creates learning data by statistically processing plural pieces of measurement data, which may include abnormal measurement data, of an energy storage device; a storage unit that stores a model learned to output a score corresponding to whether or not abnormal measurement data is included in the measurement data when the measurement data is input using the created learning data; and a detection unit that detects an abnormality or a sign of abnormality of the energy storage device based on the score output by inputting the plurality of pieces of measurement data to the model.