A Day Ahead Prediction of Solar PV Power Output Using Ensemble Neural Network

Solar PV power prediction plays a major role in the economic operation of power systems. This paper investigates the performance of various neural network frameworks in the short-term prediction of photovoltaic power output. These frameworks are the Feedforward Neural Network (FNN), the Cascade Neural Network (CNN), and the Ensemble Neural Network (ENN). In addition, the neural network models are trained using a variety of training algorithms, namely the Levenberg-Marquardt, Bayesian regularization propagation, and Scaled Conjugate Gradient. Results show that the ensemble neural network framework has a better prediction performance compared to the other methods. Furthermore, the forecasting accuracy of the Ensemble Neural Network is found to be irrelevant to cloud coverage conditions.