Author(s)

Li Zhang

Corresponding Author

Li Zhang

Abstract

Aiming at the problems of high energy consumption, complex regulation and control of building electrical system, and it is difficult for traditional methods to give consideration to both global optimization and local stability, this paper proposes an intelligent regulation and control method that integrates graph neural network (GNN) and double-layer reinforcement learning (DRL). Firstly, a spatiotemporal GNN (ST-GNN) based energy consumption prediction model is constructed, which integrates BIM topology and IoT temporal data to achieve joint modeling of multi device spatial correlation and temporal dependence, significantly improving prediction accuracy. Secondly, a double-layer control architecture of "global optimization of upper DRL+local control of lower PID" is designed, which takes into account system-level energy-saving optimization and stable operation of equipment. Furthermore, the model distillation technology is used to realize the migration of the cloud model to the edge lightweight model, which meets the real-time response requirements of 100ms. The experimental results show that this method achieves 18% energy-saving effect in real office buildings, the prediction error is reduced to 2.68 kW·h (Mae), and the average response time is 87ms, which verifies its comprehensive advantages in energy consumption prediction, intelligent regulation and real-time, and provides a feasible technical path for green intelligent buildings.

Keywords

Energy Consumption Prediction, Intelligent Regulation, Building Electrical System, Artificial Intelligence, Graph Neural Network, Reinforcement Learning