A very short-term global solar irradiance forecasting of photovoltaic generation systems using backpropagation neural network

Abstrak

Solar power plants, which are a form of renewable energy, are highly dependent on the variability of solar radiation intensity. To ensure the stability and efficiency of the system, an accurate and very short-term solar radiation intensity forecasting model is required. The objective of this study is to develop a global solar irradiance forecasting model that predicts one-hour-ahead data using a backpropagation neural network (BPNN). The data used are the results of meteorological variable measurements, including solar irradiance, air temperature, humidity, panel output power, and clarity index, collected from the solar power generation system at Surabaya State University. Training and testing were conducted using a trial-and-error approach. Performance evaluation was conducted using the MSE, RMSE, MAPE, and R² metrics. Simulation results showed that the network configuration with 15 input neurons, 25 hidden layer neurons, and 1 output neuron trained with 2000 epochs provided the best performance, with an R² value of 0.98, an average MAPE of 5.89%, the smallest RMSE of 0.04, and the smallest MSE of 0.00027. This model is capable of capturing the temporal patterns of historical data and has proven effective in predicting very short-term solar irradiance.

Identifier

• DOI: https://doi.org/10.22219/kinetik.v11i1.2449
• OpenAlex: https://openalex.org/W7131352694