Multi-agent reinforcement learning enabling dynamic pricing policy for charging station operators

The development of plug-in electric vehicles (PEVs) brings lucrative opportunities for charging station operators (CSOs). To attract more CSOs to the PEV market, provision of reasonable pricing policy is of great importance. However, dynamic environments and uncertain behavior of competitors make the pricing problem of CSOs challenging. In this paper, we focus on the dynamic pricing policy for maximizing the long-term profits of CSOs. Firstly, we propose a hierarchical framework to describe the economic association of PEV market, which is composed of smart grid, CSOs and charging stations (CSs) serving PEVs from top to bottom. Next, we leverage the Markov game to model the layer of CSOs as a competitive market. Finally, we design a dynamic pricing policy algorithm (DPPA) based on multi-agent reinforcement learning to achieve higher long-term profits of CSOs. Based on the real data of PEVs in Beijing, the experiment results show that DPPA has a significant improvement in long-term profit of CSOs, and the improvement gains increase over time. Moreover, DPPA can reduce the profit loss of CSOs effectively while involving more competitors.