Bionic cooperative load frequency control in interconnected grids: A multi-agent deep Meta reinforcement learning approach

Abstract

In the interconnected power grid operating within a performance-based frequency regulation market, uncoordinated frequency control strategies and power fluctuations in interconnection lines can intensify conflicts of interest among grid operators, leading to frequent and severe frequency fluctuations. To address these challenges and enhance grid stability, the Squid-Inspired Cooperative Load Frequency Control (SC-LFC) method is proposed. This method mimics the distributed neural decision-making observed in squids, treating each unit within an area as an independent agent. In real-time applications, each unit independently collects local frequency and status information, thereby avoiding coordination failures due to inter-area communication delays or errors. To achieve efficient coordinated control across multiple objectives and regions in complex, random interconnected power grids, the Automatic Curriculum Multi-Agent Deep Meta Actor-Critic (ACMA-DMAC) algorithm is introduced. This approach employs a hybrid curriculum learning strategy, enabling gradual learning and adaptation, which enhances the robustness and efficiency of the SC-LFC strategy. Simulations based on a four-area load frequency control model of the China Southern Grid (CSG) validate the effectiveness and superior performance of the proposed method.