Home Energy Management System based on Deep Reinforcement Learning Algorithms

Abstract

With the recent progress in smart grid applications, home energy management system has increased its importance since it allows prosumers to be active participants of the system operation. Operating the smart grid in an efficient way without having a contingency issue has become paramount. The uncertainty of the system inputs, such as renewable energy and load consumption, with the effect of dynamic user behavior, brings the necessity of a more complex control system. In this paper, we introduce three different Deep Reinforcement Learning (DRL) algorithms to minimize the operational cost in the long run and keep the battery state of charge (SoC) between the operable limits. The idea behind applying three different DRLs is to present the powerful and weak sides of the DQN, DDPG, and TD3 algorithms in terms of solving a management problem, even with the continuous state and action space for longer horizons. Experimental results show that the proposed RL algorithms can be employed to solve this and similar management problems. These show that DRL algorithms promise to solve even more complex problems with their uncertainties, but it is difficult to guarantee that they will reach an optimal solution.