Research on Machine Learning-Based Fault Diagnosis and Prediction Techniques for Power Transformers

Due to the diversity and heterogeneity of electrical equipment, the operation monitoring systems for power transformers are undergoing diversified development. By utilizing machine learning methods, this study focuses on the prediction and diagnosis of heterogeneous terminal data from power transformers to address the challenges of online monitoring. This research is of significant importance for achieving the interconnection of all components in the power system and maintaining its operational stability. The study adopts machine learning techniques to comprehensively analyze the heterogeneous terminal data of power transformers, considering seven electrical energy parameters and four environmental parameters as the main input variables for the experimental analysis. A weighted least squares support vector machine algorithm based on chemical reaction optimization is proposed for parameter prediction in power transformers. Through a detailed comparative analysis during the simulation process, with a focus on evaluating regression algorithm performance metrics, the simulation results demonstrate that the proposed algorithm can rapidly and accurately predict parameter values for future time intervals.