Fault Identification Method for Distribution Networks Based on Time–Frequency Spatial Fusion Matrix

Abstract

To address the issues of confusing local measurement point features and insufficient generalisation ability of traditional models in distribution network fault identification, this paper proposes a fault identification method integrating an optimisation mechanism and a gradient boosting model. First, based on the spatiotemporal propagation characteristics of fault signals, the method analyses the differences in propagation of travelling wave signals generated by operational disturbances and faults in the power grid. To accurately quantify the aforementioned spatiotemporal propagation laws, it extracts time–frequency domain statistical features from multiple measurement points across the entire network. These features include modal energy, central frequency, spectral entropy, frequency change rate, and sparsity index and constructs an anticonfusion feature matrix. Subsequently, the Whale Optimisation Algorithm is used to dynamically adjust the hyperparameters of XGBoost, thereby establishing the WO-XGBoost identification model. This enhances the accuracy and speed of the model during the identification process, enabling accurate identification of distribution network faults. Experimental results show the proposed method outperforms mainstream existing methods in both identification accuracy and training efficiency, offering reliable technical support for distribution network fault identification.