Multi-time scale hierarchical dispatch method for multi-region interconnection system based on multi-agent advantage actor–critic approach

In response to the issues of large-scale dispatching calculations and overly conservation of wind and solar uncertainties in the modern power systems, this paper proposes a multi-time scale hierarchical dispatch method for multi-region interconnection system. This method includes day-ahead two-stage robust optimization, intra-day multi-agent reinforcement learning rolling decision-making, and real-time distributed coordinated control. In the time dimension, a three-layer structure is constructed, ranging from day-ahead robust optimization to intra-day Multi-Agent Advantage Actor-Critic rolling optimization and then to real-time distributed sub-gradient coordination. In the spatial dimension, neighbor strategy fingerprints and electrical distance spatial discounts are introduced, and an Multi-Agent Advantage Actor-Critic control mechanism driven by dynamic spatial discounts and high-frequency fingerprints is proposed. The simulation results show that the proposed method significantly reduces the dispatching cost and achieves 5.53% curtailment of wind and solar power under multi-source uncertainty scenarios. In spatial coordination and extreme disturbance scenarios, it significantly reduces the power fluctuations of tie-lines, the number of line congestions, and voltage violations, and accelerates the frequency recovery speed, thus verifying its economic efficiency, robustness, and feasibility.