Investigation of hybrid adversarial-diffusion sample generation method of substations in district heating system

District heating system (DHS) are the largest energy-consuming component in the building energy sector. The management and operations of DHS are crucial in supporting energy efficiency and reducing carbon emissions. Existing management and operational techniques rely on sufficient load samples. However, the short operational time and equipment outages often result in insufficient load samples, challenging supporting effective management and operations. This study compares mainstream generation models of generative adversarial networks (GAN) and the denoise diffusion probabilistic model (DDPM), considering performance aspects such as accuracy, generalization, and robustness. Based on this comparison, a hybrid model for generating load samples in DHS is proposed, and this model has been validated in a DHS located in Changzhou, China. It achieves a low relative error of 1.15% in generating samples task. To validate the effectiveness of the generated samples, an indoor temperature response model is developed to calculate the indoor temperature based on load samples. This model achieves a root mean square error of 0.248 °C, effectively bridging the “external condition-load sample-indoor temperature” relationship. This research addresses the challenge of insufficient load samples in DHS and serves as a reference for addressing similar issues in energy systems.