Machine Learning Forecasting Model for Solar Energy Radiation

Renewable systems such as solar and wind are intermittent by nature. This attribute makes integrating them on a large-scale generation difficult for optimum utilization. Due to this challenge, several forecasting models have been developed to address the issue. The problems of the existing methods forecasting models are computational complexity, overfitting and low accuracy. This paper proposes a deep learning model called Long Short-Term Memory (LSTM) to forecast solar energy radiation using meteorological features. Selected hyperparameters of the proposed LSTM model are optimized with the Grid Search Cross-Validation (GridSearchCV) method. Four Machine Learning (ML) methods, Support Vector Regression (SVR), Random Forest Regression (RFR), Extreme Gradient Boosting (XGBoost) regression, and stacked RF-XGBoost, are investigated as benchmark models for the proposed LSTM-GridSearchCV model. The experimentation results revealed that the proposed method is superior to the benchmark ML model regarding accuracy and performance errors technique and capable of accurately forecasting the solar energy system. It can help the practitioner make accurate decisions on integrating renewable energy into a large-scale system.