A Hybrid Knowledge-Based and Data-Driven Method for Photovoltaic System Fault Diagnosis

This article investigates the fault diagnosis problem of the photovoltaic system. The knowledge-based method or the data-driven method may cause diagnosis error for the photovoltaic system. Thus, this paper proposes a hybrid knowledge-based and data-driven method. First, a simulation model representing the output characteristics of the photovoltaic arrays is constructed in MATLAB/Simulink. Within this model, a deep belief network (DBN) is employed as the fault diagnosis algorithm. To optimize the DBN's performance, the bat algorithm is applied to determine the ideal number of neurons in each hidden layer. The resulting optimized DBN serves as the fault diagnosis model. By collecting and normalizing the output characteristic samples under different operating conditions, the proposed model is trained and utilized for fault diagnosis in the photovoltaic arrays. Simulation results demonstrate that the proposed algorithm surpasses KNN, BPNN, and the original DBN algorithm in terms of fault diagnosis accuracy in photovoltaic arrays. It exhibits superior classification performance and is better suited for fault diagnosis in such arrays.