Multi-Microgrid Energy Trading Strategy Based on Multi-Agent Deep Deterministic Policy Gradient Algorithm

Abstract

Compared to individual microgrid, multi-microgrid (MMG) system can enhance the overall utilization of renewable energy, effectively improve the operational stability of local microgrids, and reduce the dependence on main grid. However, energy management of MMG encounters significant challenges due to the complex interaction between different microgrids. To tackle this issue, this paper introduces a non-cooperative gamebased optimal scheduling market trading model for MMG composed of various renewable energy sources, completing trade decisions while ensuring information independence. Considering the real-time changes in environmental transition functions and complex scheduling scenarios, the multi-agent deep deterministic policy gradient (MADDPG) algorithm is employed, which modifies the experience replay mechanism and Markov process of the basic deep deterministic policy gradient (DDPG) algorithm. Compared to traditional multi-microgrid system scheduling algorithms, the method presented in this paper does not require individual predictions of state variables, achieves end-to-end training from agent states to actions, and ensures the information security and autonomous decision-making of each microgrid.

Keywords

• multi-microgrid
• deep reinforcement learning
• energy trading
• multi-agent deep deterministic policy gradient
• energy scheduling
• 68t07