Enhanced artificial neural networks for simultaneous multi-parameter estimation in photovoltaic systems

Electricity has become one of the adaptable forms of energy, and the capability to convert solar energy into electricity has facilitated its usefulness. Thus, estimation of the parameters of photovoltaic systems is critical for the design and establishment of an energy system. The objective of the study is to develop an approach that can simultaneously predict large parameters of photovoltaic systems with a high degree of accuracy for energy generation industries. The study introduced radial basis function neural network (RBFN) and feedforward backpropagation artificial neural network (FFBN) algorithms for simultaneous estimation of parameters of photovoltaic systems. In this study, two enhanced artificial neural network approaches are employed to estimate the parameters of photovoltaic systems simultaneously. The two approaches are the enhanced direct artificial neural network (EDANN) and enhanced inverse artificial neural network (EIANN). The results of the enhanced artificial neural network approaches were benchmarked against the findings of the conventional artificial neural network, random forest, and multiple linear regression. The output parameters for both of the enhanced models are radiation insolation (W/m²), generated power output (kWh), total energy (kWh), module temperature (°C), direct current (A), grid current (A), and voltage (V). The input parameters for the enhanced direct artificial neural network are the computed virtual functions from the selected candidate parameters. The input parameters for the enhanced inverse artificial neural network are the computed virtual functions and the predicted output from a single-output model. The accuracy of the EDANN is found to be higher than that of the EIANN for the FFBN and RBFN algorithms. Moreover, the accuracy of the FFBN approaches was found to be higher than that of the RBFN approaches. The accuracy of the FFBN-based enhanced direct artificial neural network is 1.02, 1.07, and 1.05 times that of the enhanced inverse artificial neural network in terms of mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE), respectively.