Research on Switchgear Partial Discharge Signal Type Identification Based on Composite Neural Network

Due to the low accuracy of traditional machine learning algorithm to identify the local discharge signal type, this paper proposes a BP neural network (BPNN) identification method based on improved whale optimization algorithm (WOA)—Cubic Improved Whale Optimization Algorithm Back Propagation Neural Network (CIWOA-BP) to achieve the purpose of improving the accuracy of local discharge signal type identification in switchgear. The method improves the algorithm ability and convergence accuracy of the traditional WOA by combining adaptive inertia weights with improved convergence factors through Cubic chaotic mapping of the initial population, and the improved algorithm gives the optimal weights and thresholds to the BPNN to enhance the recognition accuracy of the model, while maintaining the generalization capability and fault tolerance of the BPNN. In this paper, three types of TEV partial discharge (PD) models of switchgear are established, the feature quantities are downscaled using wavelet soft threshold denoising and principal component analysis (PCA), and the CIWOA-BP algorithm is used to classify the discharge defect types. The results show that the recognition rates of CIWOA-BP are all above 93.3%, which is better than WOA-BP and BPNN models, proving the practicality of this recognition method.