Energy Storage in the Smart Grid: A Multi-agent Deep Reinforcement Learning Approach

This chapter introduces an energy storage system controlled by a reinforcement learning agent for smart grid households. It optimizes electricity trading in a variable tariff setting, yielding consumer savings averaging 20.91% annually without altering consumption habits. Integrated with solar panels, it offers even greater cost reductions. A multi-agent system simulation analyzes interactions between agents and identifies beneficial price–demand relationships. Moreover, it shows storage’s positive impact on the energy market for operators and consumers. Deep Q-Learning is identified as the most effective algorithm, efficiently managing high-dimensional, nonstationary, and stochastic aspects of the problem, bypassing the need for abstract modelling and deterministic rules. Furthermore, our ablation study explores various storage sizes and agent complexities.