Hybridization of Meta-Evolutionary Programming and Artificial Neural Network for predicting grid-connected photovoltaic system output

Abstract

The unpredictable weather conditions has motivated the need of predicting the output of photovoltaic (PV) system. This paper presents a Grid-Connected Photovoltaic (GCPV) system output prediction scheme using hybridization of Evolutionary Programming (EP) and Artificial Neural Network (ANN). In this study, the AC kWh output of a GCPV system was predicted using ANN based on solar irradiance (SI) and PV module temperature (MT) as the inputs. In addition, a Meta-EP was hybridized with a Multi-Layer Feedforward Neural network (MLFNN) to search for the optimal number of neurons in hidden layer, the learning rate, the momentum rate, the type of activation function and the learning algorithm during ANN training such that the root mean square (RMSE) of the prediction could be minimized. Besides Meta-EP, other variations of EP were also tested for the hybridization with MLFNN such that the proposed Meta-EP could be justified. The results showed that Meta-EP based hybrid MLFNN (HMLFNN) had produced the lowest average RMSE, the lowest standard deviation (STD) and the lowest computation time during training when compared to other EP-based HMLFNNs. Similarly, during testing, the Meta-EP based HMLFNN had also outperformed the others in producing the lowest RMSE. In the comparisons, the coefficient of determination was found to be relatively very close to unity such that a high prediction performance could be ensured.