Improved Random Forest Fault Diagnosis Method for High Voltage Circuit Breaker Based on Reconstructed Feature Matrix and Sliding Window Method

High-voltage circuit breakers (HVCB) are the key equipment for power transmission in high-voltage grids. For the problem that the traditional classification algorithm does not have high accuracy for fault diagnosis of HVCB in the case of insufficient fault data. This paper proposes a method based on reconstructing the feature matrix and sliding windows to improve the random forest algorithm. First, the Gini index is used to feedback the relative importance of all features and determine the distribution of important features, which is then used as a basis to reconstruct the feature matrix. The reconstructed single sample is then divided into multiple subsamples using the sliding window method, and all of them are used in the training of the decision tree after indicating their labels. The proposed method not only improves the diagnostic accuracy of the traditional model, but also performs better in small samples.