Efficient and precise control of district heating (DH) networks is a critical pathway for achieving energy optimization and carbon emission reduction. This study proposes a systematic approach integrating data augmentation, hybrid model forecasting, and cost optimization. First, a Generative Adversarial Network (GAN) is employed to generate scenarios from limited meteorological and operational data, constructing an expanded dataset. Based on this, a personalized load forecasting model utilizing a dynamically weighted LSTM–Prophet combination is developed. This model assigns personalized weights to each heating station to accommodate the operational requirements of different functional zones. Validated using a district heating network in Tianjin, the results indicate that with an optimal weight of w = 0.9, the average relative error for load forecasting at Heating Station566 is −0.65%. Furthermore, the K-means algorithm is used to cluster the scenario database. The resulting typical scenarios are input into the LSTM–Prophet model to obtain real-time loads for each station, and a cost optimization model based on the APSO algorithm is subsequently constructed. Evaluated using a representative day, the optimized system achieves a reduction in distribution-stage cost of approximately 270,600 RMB, with a saving rate of 38%.
Zheng et al. (Mon,) studied this question.