Improved Particle Swarm Optimization-Based Fuzzy Cerebellar Model Neural Network for Multiple Parametric Fault Diagnosis

Accurate and timely fault diagnosis of converter components is essential for ensuring the reliable operation of the entire power electronic system. This paper proposes a classification model for parametric faults in multiple components based on the fuzzy cerebellar model neural network (FCMNN) and the improved particle swarm optimization (IPSO) algorithm. First, in the early stage of the diagnostic algorithm, wavelet analysis is employed to extract features from the fault signals of the power electronic system, and the extracted features are used as inputs to the network. Then, the IPSO algorithm is applied to automatically optimize the initial parameters of the FCMNN, thereby constructing the IPSO-FCMNN fault classifier architecture. The proposed method is validated on a dual-buck inverter circuit platform. Compared with the backpropagation neural network (BPNN), the FCMNN, and the conventional particle swarm optimization-based FCMNN (PSO-FCMNN), the proposed IPSO-FCMNN fault classifier demonstrates superior diagnostic performance. More importantly, the proposed method reduces the arbitrariness associated with manual initialization in the FCMNN and decreases the possibility of the optimization algorithm in the PSO-FCMNN becoming trapped in local optima.