Reinforcement Learning Based Cost-Effective Smart Home Energy Management

Abstract

Demand Response (DR) techniques are regarded as the most economical and reliable way to smooth the load curve in context of the rising energy demand. In this paper, using Fuzzy Reasoning (FR) and Reinforcement Learning (RL), we have proposed a cost-effective strategy for residential demand response. This algorithm employs Q-learning, a reinforcement learning technique based on a reward system, to schedule shiftable/controllable loads optimally so that they are shifted from peak to off-peak hours of tariff. This reduces the overall electricity expenditure of a smart home while taking user comfort into account. FR is used for reward matrix generation. The suggested method works with one agent to operate 8 home appliances and makes use of fuzzy logic for rewards functions and a smaller number of state-action pairs to assess the action taken for a specific state. The Smart Home Energy Management System (SHEMS) demonstrates the application of the suggested DR scheme through MATLAB. The findings indicate that the cost of the electricity bill was reduced by 38.28%, showing the efficacy of the suggested strategy.