Optimizing Artificial Neural Network Configurations Using Multi-Factor Performance Ranking for Solar PV Systems

Abstract

Artificial Neural Networks (ANNs) are widely used for modelling nonlinear systems due to their adaptive learning capabilities. In this work, different combinations of ANN functions such as transfer functions, learning rules, and training algorithms are examined to determine their impact on prediction accuracy for solar photovoltaic (PV) modules. A dataset comprising input parameters including maximum power (Pm), open-circuit voltage (Voc), short-circuit current (Isc), irradiance (Ir), temperature (T), and fill factor (FF) from 34 different PV modules is used to predict output parameters: current (Im) and voltage (Vm) at peak power. We've developed a total of 84 ANN models by bringing together a variety of functions. Each model is trained using mean squared error (MSE), along with a set number of epochs and performance indicators. To compare the models effectively, we use a weighted ranking system. This approach helps to pinpoint the best ANN setup for accurately modeling how photovoltaic output behaves.