Automatic Load Allocation in Power Dispatch Master Stations Using Multi-Agent Deep Deterministic Policy Gradient Algorithms

Abstract

Automatic load allocation in power dispatch master stations has become increasingly challenging due to the uncertainty of renewable generation and the complexity of multi-regional power coordination. This study proposes an automatic load allocation framework based on a Multi-Agent Deep Deterministic Policy Gradient (MA-DDPG) algorithm to improve real-time dispatch efficiency and system adaptability. A distributed multi-agent architecture is established in which each dispatch region operates as an autonomous agent with coordinated communication and decision-making capabilities. By incorporating stochastic models for wind and photovoltaic generation together with a multi-objective optimization strategy that simultaneously minimizes dispatch cost, renewable energy curtailment, and load fluctuation, the proposed approach achieves coordinated global optimization under dynamic operating conditions. Furthermore, adaptive communication weighting and prioritized experience replay are introduced to enhance cooperation efficiency and accelerate policy convergence. Experimental results demonstrate that the improved MA-DDPG framework significantly enhances training efficiency, reduces operational costs, and strengthens load balancing robustness while maintaining stable system performance under renewable and demand uncertainties. The proposed methodology also provides a reliable optimization paradigm for communication-intensive smart grids, where electromagnetic information transmission and distributed sensing infrastructures support real-time coordination and intelligent energy management.