Research on prediction and optimization method of building energy efficiency based on BIM and graph neural network

With the acceleration of urbanization, the problem of building energy consumption has become increasingly prominent. Traditional building energy efficiency management methods have some limitations, such as low data utilization rate, insufficient prediction accuracy and lack of dynamic optimization ability. Building information model (BIM) technology can provide the integration and sharing of building life cycle data, while graph neural network (GNN) is good at processing graph structure data and capturing the spatial topological relationship inside buildings. In this paper, a fusion framework combining BIM and GNN is proposed. By constructing Spatio-Temporal Graph Neural Network (ST-GNN) model and integrating multi-source data to realize dynamic energy efficiency prediction, a real-time optimization strategy based on model predictive control (MPC) is designed. The experimental results show that the ST-GNN model significantly outperforms traditional physical models and single machine learning methods in energy consumption prediction accuracy, with a root mean square error (RMSE) of 15.2 kWh, a mean absolute percentage error (MAPE) of 7.9%, and a coefficient of determination (R²) reaching 0.94. In terms of energy optimization, the ST-GNN combined with MPC strategy achieves an energy savings rate of 18.6% and a comfort compliance rate of 93.4%, with control instructions updated every 15 minutes. Ablation experiments confirm the critical role of the spatio-temporal attention mechanism and BIM data in enhancing model performance. This study provides a new and efficient method for building energy efficiency management, offering significant theoretical and practical application value.