Predicting occupant energy consumption in different indoor layout configurations using a hybrid agent-based modeling and machine learning approach

Accurately predicting occupant energy consumption in buildings is essential for optimizing energy management and promoting sustainability. However, gathering reliable stochastic data on occupant energy consumption poses significant challenges. This research proposes a hybrid approach that integrates Agent-Based Modeling (ABM), System Dynamics (SD), Building Information Modeling (BIM), and Machine Learning (ML) techniques to predict energy consumption in different indoor layout configurations, including rectangular, square, and compound shapes. Initially, the hybrid model (ABM-SD-BIM) focuses on generating a comprehensive and precise dataset. Using this dataset, various ML models are developed to predict energy consumption. The results demonstrate that the ML model outperforms earlier ML models in terms of mean squared error (MSE) and root mean squared error (RMSE), indicating improved prediction accuracy. Specifically, for Layout 3, representing a compound shape configuration, the ML model achieves an MSE of 0.03 and an RMSE of 0.17. Furthermore, the ML model exhibits a high R2 score of 0.92, indicating a good fit to the data. Comparative analysis of different ML models reveals that LightGBM performs the best, with the lowest MSE and RMSE values for the compound shape configuration. On the other hand, XGBoost, RF, and DT models exhibit higher MSE and RMSE values, indicating relatively higher prediction errors. These findings underscore the effectiveness of the hybrid approach, particularly in compound shape configurations, for accurately predicting energy consumption in various indoor layouts.