A combined neural network model for forecasting solar energy from multiple weather parameters

The article aims to propose a combined neural network model to forecast the solar energy of a solar power plant. Predicting solar energy is extremely important in planning, investing in electricity development, and operating the power system. Solar power integration in electrical grids is complicated due to its dependence on volatile weather conditions. Artificial neural networks have been employed in recent years to solve the problem. Two well-known models, the Nonlinear autoregressive exogenous model (NARX) and the Long-short term memory (LSTM), are suitable methods for determining photovoltaic (PV) output forecasting. This article proposes a two-stage model combining the NARX and LSTM to forecast the output energy of a solar power plant. The proposed model has been evaluated and compared with NARX and LSTM. The data have been collected at Son My 3.1 Solar Power Plant project from the monitoring system with electrical power, solar irradiance, wind speed, and wind direction, from July 1, 2019, to June 30, 2020. A general score based on three error functions, Mean squared error, Mean absolute error, and R-squared, is the indicator to investigate the forecasting efficiency of the methods. The numerical results show that the combined NARX-LSTM model is better than other methods in this study.