A stacking ensemble model for forecasting solar photovoltaic power generation

Solar energy is an environmentally friendly and renewable energy source. In recent years, solar power plants have a large share in electricity production around the world. The use of environmentally friendly resources such as solar energy in electricity production plays a critical role in combating climate change. The production forecast of solar power plants is used to balance energy supply and demand. Highly accurate forecasts ensure the efficient operation of the energy grid, which reduces energy costs while increasing the level of energy security. In this study, we focused on the forecasting of the generation of a solar PV plant located in Istanbul, Turkey. For solar energy production, Support Vector Machine, one of the machine learning methods, and Long Short-Term Memory, one of the deep learning methods, are chosen and future forecasts are obtained with individual models. Moreover, a stacking ensemble model that includes these two methods and where linear regression is the meta-learner is proposed to increase the forecast accuracy. It is concluded that the proposed model outperformed the individual models, providing a high rate of improvement in MAE and MSE and approximately 25% improvement in msMAPE.