Energy Management Strategy for Smart Homes Based on Deep Reinforcement Learning

In residential multi-energy systems, energy management strategies enable the coordinated and cascade utilization of diverse energy sources, which reduces energy costs for households. However, energy management algorithms suffer from the problems of model uncertainty, parameter uncertainty, temporal coupling and multi-energy coupling. To address these problems, this paper proposes an energy management algorithm for a residential multi-energy system based on the deep reinforcement learning. The proposed algorithm simultaneously considers multi-energy time-shifting and cascade utilization to minimize energy costs and maintain indoor temperatures. In the proposed algorithm, an adaptive additional reward method is introduced to define the priority of different energy sources, which facilitates the exploration of the action space to achieve unconstrained energy cascade utilization. Meanwhile, the adaptive additional reward method exploits the additional reward selectively to avoid suboptimal results. Comparatives simulations demonstrate that the proposed algorithm achieves 10.32%/12.58% and 3.14%/9.75% lower daily energy costs in the low-temperature day and the high-temperature day. Furthermore, the proposed algorithm has superior performance in energy cascade utilization.