Solar Power Forecasting for Existing Solar Power Plants of Chhattisgarh Region Using Extreme Gradient Boosting Algorithm
摘要
Solar power forecasting involves the estimation of the amount of solar energy production for a specific time interval. This approach helps to manage the variability in solar energy availability and supports its seamless integration into power grids. This study emphasizes short-term solar power forecasting using the Extreme Gradient Boosting algorithm (XGBoost), a powerful machine learning method recognized for its precision and efficiency. The objective is to forecast solar power generation on a day-ahead basis for an existing solar power plant located at Sahaspur, Khairagarh, Chhattisgarh, India. The historical dataset is obtained from the Chhattisgarh State Power Distribution Company Limited, Raipur, which includes active power generation in kilowatts spanning from 25th December 2023 to 30th November 2024. The dataset is structured so that active power data from the previous seven days is used as input variables to forecast day-ahead solar power, with the target variable representing the active power for the current day. In the training, data from 25th December 2023 to 30th June 2024 was utilized, while testing data includes data from 1st July to 30th November 2024. Testing was performed on both monthly and daily basis at every 15 min to observe detailed performance. To assess the model accuracy and reliability, three commonly used metrics such as MAE, RMSE and R2 were employed, providing clear insights into forecasting performance. A key extension of this study was validating the trained model on entirely unseen datasets from other solar power plants (SPPs) without retraining, demonstrating strong generalization. The model achieved MAE/RMSE/R2 of 29.11 KW/53.52 KW/0.929 for a 1 MW plant at Mahasamund, 37.68 KW/73.39 KW/0.9819 for a 1.4 MW plant at Barbarik Tie Up Pvt. Ltd., Raipur, and 9.96 KW/14.81 KW/0.8702 for a 199.9 KW rooftop plant at NIT Raipur. These results confirm the robustness of the XGBOOST-based framework across diverse solar plant capacities and conditions, highlighting its practical application.