Reinforcement Learning-based Energy Trading for Microgrids

TL;DR

This paper introduces a reinforcement learning approach using deep Q-networks for energy trading among microgrids, optimizing utility and reducing reliance on power plants amid renewable energy variability.

Contribution

It formulates a microgrid energy trading game, derives Nash equilibrium conditions, and applies deep reinforcement learning to enhance trading efficiency in large-scale microgrid networks.

Findings

Significantly reduces power plant schedules.

Improves microgrid utility compared to benchmarks.

Effectively manages energy trading with wind generation data.

Abstract

With the time-varying renewable energy generation and power demand, microgrids (MGs) exchange energy in smart grids to reduce their dependence on power plants. In this paper, we formulate an MG energy trading game, in which each MG trades energy according to the predicted renewable energy generation and local energy demand, the current battery level, and the energy trading history. The Nash quilibrium (NE) of the game is provided, revealing the conditions under which the local energy generation satisfies the energy demand of the MG and providing the performance bound of the energy trading scheme. We propose a reinforcement learning based MG energy trading scheme that applies the deep Q-network (DQN) to improve the utility of the MG for the case with a large number of the connected MGs. Simulations are performed for the MGs with wind generation that are aware of the electricity prices and the historic energy trading, showing that this scheme significantly reduces the average power plant schedules and improves the utility of the MG compared with the benchmark strategy.