Application of reinforcement learning for energy consumption optimization of district heating system

Abstract

Heating residential spaces consumed 64 percent of total household energy consumption in Finland. Considering the heat transfer and time delay in the district heating system, the calculation of setpoints of supply temperature requires a comprehensive understanding of the real system, and experienced operators need to manually determine the setpoints. To save energy, a more effective and accurate method is needed for setpoints calculation. In this paper, a reinforcement learning based method is proposed. Through interacting with an Apros-based simulation model, the agents learn to calculate supply temperature parallelly for lowering energy costs. Simulation results show that the proposed method outperforms the existing method and has the potential to address the problem in real factories.