Improving energy management practices through accurate building energy consumption prediction: analyzing the performance of LightGBM, RF, and XGBoost models with advanced optimization strategies
摘要
The building sector has been listed among the largest energy consumers in the world, impelled by factors such as increasing population, rising demands for comfort, and climate change. Effective energy consumption forecasting is important for improving energy management, optimizing system operations, and reducing energy waste. In this study, advanced machine learning models—light gradient boosting machine (LightGBM), random forest, and extreme gradient boosting (XGBoost), are studied for application with the aid of hyperparameter optimization methods—Satin Bowerbird Optimizer (SBO) and Moth Flame Optimization. The dataset belongs to two buildings in Richland, WA, between 2009 and 2011 for developing individual and hybrid models. The results indicate that the LightGBM-SBO hybrid model demonstrated superior performance, achieving an R2 value of 0.9148, a mean absolute error of 2.42, and a root mean square error of 5.26 on the testing dataset. This study has been presented as the main novelty in integrating advanced optimization algorithms with machine learning models to substantially improve forecasting accuracy and efficiency. These results will help in the development of more resilient practices of energy management that will enable energy providers to optimally distribute loads and help in the prevention of failures, ensuring a stable system of energy supply.