A Hybrid Algorithm for Short-Term Solar Power Prediction—Sunshine State Case Study

Abstract

The growing rate of the integration of photovoltaic (PV) sites into the structure of power systems makes the task of solar power prediction more important in order to control the power quality and improve the reliability of system. In this paper, a hybrid forecasting algorithm is proposed for hour-ahead solar power prediction. A combination of gradient-descent optimization and meta-heuristic optimization approaches are designed in the structure of the presented model to take into account the prediction accuracy as well as the computational burden. At the first step, the gradient-descent optimization technique is employed to provide the initial parameters of a feedforward artificial neural network (ANN). At the next step, the meta-heuristic optimization model, called shuffled frog leaping algorithm (SFLA), is developed to search for the optimal set of parameters of ANN using the initial individuals found by the gradient-descent optimization. Then, the identified parameters by the customized SFLA will be employed by the ANN for short-term solar power prediction. The performance of the proposed forecasting algorithm is demonstrated on the solar power data of three simulated PV sites in Florida for 2006.