Comparing multilayer perceptron and probabilistic neural network for PV systems fault detection

This work introduces the development of a fault detection method for photovoltaic (PV) systems using artificial neural networks (ANN). The faults identified by the method are short-circuited modules and disconnected strings. This research's novel part is its adaptability as a long-term dataset has been used in the ANN training and validation phase and also examined situations considering datasets contaminated with random noise. It makes the method suitable for any photovoltaic power plant, also does not require long datasets from pre-existing systems or installing new sensors. The proposed method comprises two unique algorithms for PV fault detection, a Multilayer Perceptron, and a Probabilistic Neural Network. The research method used modeling, simulation, and experiment data since both algorithms were trained using simulated datasets and tested through experimental data from two different photovoltaic systems. Even though the training dataset includes noisy situations, the results indicated a superior precision for the Multilayer Perceptron neural network. The findings showed a maximum accuracy of 99.1% in detecting short-circuited modules and 100% in detecting disconnected strings.