A model predictive control method for system-wide optimization of district heating systems

Systematic management and regulation of District Heating Systems (DHSs) are key to achieving supply-demand balance and reducing energy consumption. Achieving accurate load forecasting and coordinated control across multiple heat exchange stations remains a significant challenge. This study proposes a system-wide optimization Model predictive control (MPC) method for DHSs. Based on long short-term memory (LSTM) networks and an attention mechanism (Attention) with Bayesian optimization (BO) for automatic hyperparameter tuning, a BO-Attention-LSTM (BO-A-LSTM) prediction model was established for heat exchange station heat load and secondary return water temperature, effectively improving the prediction accuracy. A system-wide optimization MPC algorithm for DHSs was designed, with an objective function that considers both heating quality and energy consumption. Particle swarm optimization (PSO) algorithm was adopted to optimize the operating parameters of heat sources, primary network circulation pumps, and primary network electric control valves at each heat exchange station, achieving coordinated control of multiple devices across different levels. Taking a typical DHS in Tianjin as the research object, TRNSYS simulation was conducted. Energy-saving effect analysis shows that the MPC strategy significantly optimized the operation of gas-fired hot water boilers, primary network water pumps, and heat exchange station control valves. Compared with the baseline control strategy, the system-wide MPC optimization reduced the total electricity consumption of the DHS to 13,448.92 kWh and the total heat consumption to 6090.47 GJ, achieving an electricity saving rate of 3.68% and a heat saving rate of 5.76%.