An Effective Hybrid Symbolic Regression–Deep Multilayer Perceptron Technique for PV Power Forecasting

The integration of Photovoltaic (PV) systems requires the implementation potential PV power forecasting techniques to deal with high intermittency weather parameters. In prediction process, Genetic Programming (GP) based on Symbolic Regression (SR) model has a widespread deployment since it provides an effective solution for nonlinear problems. However, during training SR models might miss optimal solutions due large search space leaf generations. This paper proposes novel hybrid that combines and Deep Multi-Layer Perceptron (MLP) one-month-ahead forecasting. A case study analysis using real Australian dataset was conducted, where employed input features were solar irradiation historical data. main contribution proposed SR-MLP algorithm are as follows: (1) speed significantly improved by eliminating unimportant inputs feature selection process performed Extreme Boosting Elastic Net techniques; (2) hyperparameters preserved throughout testing phases; (3) made use reduced number layers neurons while guaranteeing accuracy; (4) iterations reduced. presented simulation results demonstrate higher accuracy (reductions more than 20% Root Mean Square Error (RMSE) 30 % Absolute (MAE) in addition improvement R2 evaluation metric) robustness (preventing from converging local minima help ANN branch) compared individual MLP models.