Prediction of photovoltaic panel power outputs using artificial neural networks and comparison with heuristic algorithms

The prediction of power outputs generated from photovoltaic (PV) systems at different times is necessary for reliable and economical for use of solar panels. The prediction of the power output is also very important in terms of factors such as installation of solar panels, guidance of electricity companies, energy management and distribution. Determination of optimum solar panel positions and angles, providing energy productivity to maximize production capacity in a short time period is the most time consuming job for regulations for a companies. Also, adaptation of panels increases costs. Therefore, new and healthy prediction methods have a great importance to minimize these work force costs. In this study, Artificial Neural Network (ANN) model learned by heuristic algorithms are used for the prediction of power outputs obtained from PV panels monthly. Particle Swarm Optimization (PSO), Back-Propagation (BP), Clonal Selection Algorithm (CSA) are used to train ANN to predict six different PV panel located in different angles from 10 to 60 degrees. Three different popular evaluation methods which are called mean absolute percentage error (MAPE), root mean square error (RMSE), varyans (R^2) used to do comparison. According to examination of verification results, PSO is almost most successful algorithm as a training method when it is compared with BP and CSA. It is seen for the some of the results belong to a few months that BP is slightly better than PSO.