Multiagent Reinforcement Learning for Community Energy Management to Mitigate Peak Rebounds Under Renewable Energy Uncertainty

Abstract

Price-based demand response (DR) can aid power grid management, but an uncoordinated response may lead to peak rebounds during low-price periods. This article proposes a community energy management system based on multiagent reinforcement learning. The scheme consists of a community aggregator that optimizes the total community electricity cost for multiple residential users. A home requires energy management for home appliances, electric vehicles, energy storage systems, and renewable energy generation. The appliance scheduling problem is decomposed into smaller sequential decision problems that are easier to solve. Renewable generation is predicted and used to mitigate the influence of energy generation uncertainty. As indicated in numerical analyses, the proposed approach can handle the uncertainty in renewable energy and leads to more economical energy usage relative to existing energy management methods. The method outperforms conventional algorithms, such as centralized mixed-integer nonlinear programming and genetic algorithm-based optimization, in terms of mitigating peak rebounds and addressing the uncertainty of renewable energy generation.