Cloud–Edge Cooperative Load Frequency Control for Isolated Microgrid Using Emergent Computation-Based Large-Scale Meta-Machine Learning
Jiawen Li, Haoyang Cui
Abstract
In the isolated microgrid, each unit is managed by different operators, and their interests conflict greatly. To balance the interests of different stakeholders in the electric grid sector, this article proposes an edge–cloud cooperative data-driven load frequency control (ECCDD-LFC) method for isolated microgrids, which can deliver improved frequency control performance and reduces regulation cost. This method replaces the LFC controller in the conventional microgrid control center with a cloud-based agent and sets each unit as an independent-decision edge-based agent. It enables the units to make decisions independently and realize cloud–edge cooperative control. In addition, an emergent computation large-scale multiagent deep meta-deterministic policy gradient (ECLSMA-DMDPG) is proposed in this article. In this algorithm, a centralized training and decentralized execution policy are used to achieve the collaboration of multiple agents, a meta-learning technique is used to improve the robustness of the algorithm, and emergent computation is used to produce high-value samples for a higher performance policy. The effectiveness of the proposed method is demonstrated in a simulation of the Zhuzhou isolated microgrid of the China Southern Grid (CSG).