Resilient energy management of a multi-energy building under low-temperature district heating: A deep reinforcement learning approach

The corrective control of a building-level multi-energy system (MES) for emergency load shedding is essential to optimize the operating cost after contingency. For a Danish case, the heating devices in the building are connected to a developing low-temperature district heating (LTDH) system and operated under a heat market. Due to the coupling between the electrical power and heating system, an electricity outage can be propagated to the heating network, and heat prices as well as tariffs can impact the MES operating cost. In the previous studies, only electrical load shedding is modeled, while the impact of electricity outages on heating system operation and heat load control is ignored. On the other hand, the problem is traditionally solved by model-based optimization methods which are highly nonconvex leading to high computing complexity. Moreover, operating uncertainties can lead to infeasible solutions. To address these challenges, this paper proposes a deep reinforcement learning-based corrective control method for the resilient energy management of a building-level MES. In the method, the proximal policy optimization algorithm is applied, where multiple uncertainties, system dynamics, and operating constraints are considered. A case study of a real-life residential building connected to the LTDH system in Denmark is carried out, where electricity outages are simulated. The results verify the performance of the proposed method in achieving resilient energy management of the MES.