Energy Consumption Prediction for a Recreation Facility Using Data-Driven Techniques

Leisure centres are multifunctional buildings that have irregular energy consumption patterns and consume more energy compared to most building types. However, they have little representation in building performance energy prediction literature. This work presents an energy consumption prediction effort for a leisure centre using data-driven techniques, namely Light-gbm, support vector regression and multi-linear regression models. Climatic and energy use data collected over sixteen months were pre-processed, normalized and split into training and testing sets for regression analysis. The results showed that the ensemble-based Lightgbm model had superior performance in a multi-input prediction setting. The support vector regression model and multi-linear regression had a marginal difference between themselves at the prediction task. The MAE, RMSE and R² evaluation metrics ranged from good to very good among the created models. The previous energy consumption observation is determined as the essential variable for energy consumption at this multi-functional building type. The developed predictive models can be an alternative method for the better attainment of efficient energy management.