Diffusion model-driven multi-objective collaborative optimization for building energy management using graph neural networks

To solve the high-dimensional nonlinear multi-objective energy optimization in building clusters, this paper proposes a generative framework driven by conditional diffusion models integrated with graph neural networks. The joint equipment scheduling is transformed into a conditional denoising generation process, where a multi-layer GNN encodes spatial-topological dependencies and grid boundaries to guide the reverse trajectory. Validated via the open-source Building Data Genome Project 2 dataset and the high-fidelity EnergyPlus simulation platform, the method is evaluated against NSGA-II, MOPSO, and Deep Reinforcement Learning baselines. Results demonstrate a 28.6% average energy-saving rate, a 38.5% peak-load reduction in summer, and a hypervolume indicator of 0.95. The framework scales stably up to 50 buildings with a 31.2% energy-saving rate.