A new prediction model for solar irradiance using ant colony optimization and neural network

In this paper, a new prediction model for solar irradiance has been proposed using ant colony optimization (ACO) and neural network (NN), called as ACOSIP. In ACOSIP, the most salient climatological features are selected in order to enhance the solar irradiance prediction (SIP) accuracy. To implement such idea, ACO search technique utilizes the advantages of the combined activities of the features by considering the correlation information among the features and the outcome of NN. Thus, ACOSIP introduces the wrapper and filter approaches in its feature selection process. To make an effective ACO search, two sets of new rules have been designed for pheromone update and heuristic information measurement. To evaluate the performance of ACOSIP, 12 solar irradiance data samples in between the year of 2000-2013 were collected from Bangladesh Meteorological Department (BMD). Experimental results show that ACOSIP can select six most salient features easily with increasing the prediction accuracy, which are longitude, latitude, day light hour, max temp., min temp., and humidity. In addition, the averaged prediction accuracy of ACOSIP for 35 stations of BMD in testing case is 99.74% including the MAPE of 0.26%. The proposed ACOSIP also represents high correlation of 99.93% in between the actual and forecasted data.