Day-ahead Photovoltaic Power Production Forecasting Using a Hybrid Artificial Neural Network Model Integrated with Metaheuristic Algorithms

The escalating global energy demands and the environmental repercussions of fossil fuel utilization have given rise to a marked increase in the level of interest in renewable energy sources. Solar energy, in particular, is distinguished by its abundance and minimal environmental impact. This study sets out to compare three distinct hybrid models that are designed to enhance the forecasting accuracy of daily photovoltaic power prediction: JAYA-ANN, GA-ANN and PSO-ANN. The models were developed and tested using historical data on PV power output, including air temperature, PM10 levels, and solar irradiance. The study’s findings indicated that the JAYA-ANN hybrid model exhibited superior performance, with a Mean Absolute Percentage Error (MAPE) of 7.38% and a Root Mean Squared Error (RMSE) of 681.71 kW for the test subset. The JAYA-ANN model demonstrated superior performance in comparison to both GA-ANN and PSO-ANN models. On the basis of the entire dataset, the JAYA-ANN model exhibited the highest level of prediction accuracy, with an MAPE of 11.59% and an RMSE of 413.91 kW. The study confirms that the JAYA-ANN hybrid model serves as an effective tool for photovoltaic power estimation. Beyond this, it offers noteworthy opportunities to advance the integration of solar resources into the energy sector while maintaining grid stability through enhanced forecasting accuracy.