Using PSO-GA algorithm for training artificial neural network to forecast solar space heating system parameters

An Artificial Neural Network (ANN) model based on PSO-GA optimization algorithm is applied to predict a Solar Space Heating System (SSHS) performance. An experimental research is conducted into the SSHS equipped with a Parabolic Through Collector (PTC). A number of influential factors such as I, T-a, T-2 and T-w are considered to validate the ANN results. The proposed PSO-GA algorithm is used to identify a complex non-linear relationship between input and output parameters of the SSHS, and to obtain the optimized estimating ANN model. To show the accuracy of the PSO-GA model in training ANN, its results are compared with those of two highly powerful optimization algorithms, namely High Exploration Particle Swarm Optimization (HEPSO) and Team Game Algorithm (TGA), based on some evaluating criteria such as Mean Absolute Percentage Error (MAPE), Variance Accounted For (VAF), the coefficient of determination (R-2) and Root Mean Square Error (RMSE). Results show the reliability of PSO-GA-ANN with highest R-2 and RMSE. Afterwards, testing phases of the PSO-GA algorithm are done based on its parameters to obtain the best model of the neural network. Eventually, comparing the proposed model to Multi-Layer Perception Artificial Neural Network (MLPANN) exhibits its higher ability to estimate outputs when more accurate results are required.