Multi-Agent Smart Control of Virtual Power Plant Energy Storage in Active Distribution Network

The current distribution grid renewable energy penetration rate is increasing, and in order to improve the regulation capacity of the grid, the distribution grid is in urgent need of peak and frequency regulation resources. Energy storage systems (ESS), as a flexible energy regulation resource, is now being used on a large scale at all levels of the grid as its cost decreases and technology matures. Virtual power plant (VPP) taking ESS into consideration can effectively regulate internally distributed energy and externally present the characteristics of a power plant, which can be used as an emerging resource for peak and frequency regulation. However, the distribution network has the problem of complex network structure and poor objective measurement, which is difficult for traditional centralized control to handle. A multi-intelligent reinforcement learning algorithm with Centralized Training Decentralized Execution (CTDE) architecture is used to solve the control of multiple ESS systems in active distribution network VPP, while considering the operational loss cost of ESS and the uncertainty of renewable energy sources (RES) prediction. The results show that the proposed MADDPG-based smart control method can effectively achieve the response of ESS to electricity price with less deviation of RES assessment and ESS life loss cost.