Transformer Fault Forecasting Using Dissolved Gas Rate of Change
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transformer fault detection methods using static alarm thresholds for dissolved gas analysis often result in false positives, especially in grids with distributed energy resources and varying loads, leading to unnecessary maintenance and potential transformer shutdowns.
Innovation Solution
A method that uses machine learning models like Prophet and DeepAR to forecast dissolved gas concentrations and their rate of change, allowing for adaptive alarm thresholds and proactive maintenance by clustering similar transformers and selecting the best model for each cluster to predict future faults accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If static alarm thresholds are used for dissolved gas analysis, then the detection method is simple and easy to implement, but false positives increase leading to unnecessary maintenance
Solution Approach 1:
The patent transitions from static alarm thresholds to dynamic, adaptive thresholds that automatically adjust based on transformer-specific baseline characteristics and operational conditions. The system learns normal gas generation patterns for each transformer and adapts thresholds accordingly, reducing false positives while maintaining ease of operation through automated adaptation.
Solution Approach 2:
The system changes the parameter of alarm thresholds from fixed static values to dynamic values that evolve based on learned transformer behavior patterns. By continuously updating baseline characteristics and adapting thresholds to match actual operational conditions, the system improves detection accuracy without complicating implementation.
2Reliability
If adaptive alarm thresholds with machine learning models are used, then false positives are reduced and detection accuracy improves, but system complexity increases
Solution Approach 1:
The system performs self-service by automatically learning transformer-specific baseline characteristics and adapting alarm thresholds without requiring manual configuration or expert intervention. The machine learning models autonomously train on historical data and adjust parameters, reducing the perceived complexity for users while maintaining high detection accuracy.
Solution Approach 2:
The system performs preliminary learning and adaptation during normal operation before faults occur, building baseline characteristics and training models in advance. This preliminary action prepares the system for accurate fault detection without adding complexity during critical monitoring phases, as the heavy computational work is distributed over time.
3Productivity
If transformer-specific baseline characteristics are learned and adapted, then maintenance optimization improves, but data processing requirements and computational resources increase
Solution Approach 1:
The patent segments the transformer fleet into clusters based on similar characteristics and operational patterns. By applying machine learning models at the cluster level rather than individual transformer level, the system reduces computational resources while maintaining personalized baseline characteristics for each transformer within its cluster, optimizing maintenance schedules efficiently.
Solution Approach 2:
The system develops universal machine learning models that can be applied across multiple transformers within clusters, making the computational work reusable. Once a model learns patterns for a cluster, it serves multiple transformers simultaneously, reducing overall computational resource requirements while maintaining personalized maintenance optimization for each unit.
Data Source
AI summary
A method for using forecasting power transformer faults may include receiving dissolved gas data of power transformers; receiving features of the power transformers; generating, based on the features and the dissolved gas data, clusters of transformers, wherein each respective cluster of the clusters comprises transformers exhibiting similarities to each other; selecting, for each respective cluster and respective gas of the first dissolved gas data, from among machine learning models trained to minimize a difference between a forecast gas concentration and an actual gas concentration, a machine learning model; generating, using a selected machine learning model, a forecasted concentration of the respective gas of the respective cluster; estimating, using forecasted gas concentration, a ROC of the respective gas of the respective cluster; predicting, based on a comparison of the ROC to an ROC alarm threshold, a future fault of a transformer; and generating an alert indicating the transformer maintenance is required.