Photovoltaic System Fault Diagnosis Using Feature Engineering and Bayesian-Optimized CNN

This paper presents a novel data-driven method to enhance the efficiency of fault diagnosis methods for photovoltaic (PV) arrays. Optimized deep learning algorithms with feature engineering techniques are integrated in this study to diagnose single and hybrid faults in PV arrays. First, a simulation model generating the PV array’s current-voltage (I-V) curve characteristics under various operational conditions is constructed using MATLAB/Simulink. Then, a new feature engineering is proposed to normalize the I-V curve characteristics to overcome the similarity of curves in various operating conditions. After that, a one-dimensional convolutional neural network (1D-CNN) model is suggested for the classification issues. To optimize the 1D-CNN model performance, a Bayesian optimization algorithm is used to choose the ideal number of layers and best settings for this model. Finally, the optimized 1D-CNN model is trained using the normalized features. Experimental results demonstrated that the optimized 1D-CNN model accurately identifies and diagnoses single and hybrid faults within PV arrays. Additionally, it is shown that the proposed model outperformed other machine learning models regarding diagnostic accuracy.