Reinforcement learning control strategy for differential pressure setpoint in large-scale multi-source looped district cooling system

Large-scale multi-source looped district cooling (MLDC) systems are expected to be a promising solution to support decarbonization goals and combat climate change owing to their advanced structure. Differential pressure control is one of the most common methods to improve the energy and hydraulic performance of MLDC systems. Traditional rule-based control (RBC) and model-based control may not be appropriate for large-scale systems because of the need for expert knowledge and accurate models. Reinforcement learning control (RLC) has attracted considerable research attention owing to its efficiency and flexibility. However, very little is known about RLC in large-scale heating, ventilation, and air conditioning (HVAC) systems and complex issues. Therefore, this study employs a reinforcement learning technique to optimize the differential pressure setpoints of multiple cold sources, which can achieve energy savings and fulfill hydraulic head demands simultaneously. In this study, a Modelica model of a largescale MLDC system in Beijing was developed as a virtual environment. The Modelica-Python cosimulation testbed for RLC was then implemented. The results show that RLC can save up to 12.99% of the annual distribution energy consumption while achieving a good hydraulic performance. As an information base for a variety of stakeholders, this study offers a reinforcement learning solution that can improve the operating performance of large-scale HVAC systems. (c) 2023 Elsevier B.V. All rights reserved.