Cooperative Dispatch of Renewable-Penetrated Microgrids Alliances Using Risk-Sensitive Reinforcement Learning
Ziqing Zhu, Xiang Gao, Siqi Bu, Ka Wing Chan, Bin Zhou, Shiwei Xia
Abstract
The integration of individual microgrids (MGs) into Microgrid Alliances (MGAs) significantly improves the reliability and flexibility of energy supply. The dispatch of MGAs is the key challenge to ensure the secure and economic operation of the distribution network. Currently, there is a lack of coordination mechanism that aligns the individual MGs' objectives with the overall welfare of the alliance. In addition, current optimization method cannot simultaneously achieve requirements of MGAs' dispatch, including fast computation speed, scalability, foresight-seeing capability, and risk mitigation against uncertainty due to high penetration of renewable distributed energy resources. In this paper, a cooperation mechanism for MGs in the MGA is proposed to harmonize MGs' own profit and the global profit of the MGA, with the guarantee of fairness. Aligned with this mechanism, a novel Risk-Sensitive Trust Region Policy Optimization (RS-TRPO), as a risk-averse multi-agent reinforcement learning algorithm, is proposed to help MGs to optimize their own dispatch strategy. This algorithm tackles the deficiencies of conventional methods, enabling the distributed, fast-speed, and foresight-seeing dispatch of MGs in a scalable manner, while considering the uncertain risks. In particular, the optimality of this algorithm is theoretically guaranteed. The outstanding computational performance is demonstrated in comparison with conventional algorithms in a modified IEEE 30-Bus Test System with 4 MGs.