Fault detection method for photovoltaic array using LSTM network

The faults in the photovoltaic (PV) array severely hinder power generation and system operation. The detection of PV array faults becomes a complex assignment using the conventional protection system as the various fault cases and normal operating conditions manifest the same characteristics at certain conditions. This paper proposes a new method to detect faults in PV arrays using an LSTM network. Initially, an analysis of the fault characteristics of the 3.2 kW PV array under various fault conditions is performed to identify the input parameters for the LSTM network, enabling it to detect short circuit faults (SCFs), line-ground faults (LGFs), and hot spot faults (HSFs) in the PV array. It is found that the PV array measurements of current (I), voltage (V), power (P), rate of change of current, and rate of change of voltage to time (dI/dt and dV/dt), and change in power to change in voltage and change in current (dP/dV, and dP/dI) provides the useful characteristics of fault for classification of faults. The proposed LSTM network is trained with the following measurements of the PV array: I, V, P, dI/dt, dV/dt, dP/dV, and dP/dI. The effectiveness of the proposed method is validated with the testing dataset and the performance indices derived from the confusion matrix. Further, the accuracy of the proposed method is contrasted with other intelligent techniques like artificial neural network (ANN), support vector machine (SVM), and k-nearest neighbor (kNN). The outcomes disclose that the proposed method achieves better results than alternative techniques. The modeling of the PV array, training, and testing of the LSTM networks are performed on the MATLAB-Simulink platform.