MATD3-based Multi-Agent Hydraulic Support Straightening Control with Cooperative Game Theory

Hydraulic Support Straightening control in intelligent coal mining systems faces significant challenges due to complex geological conditions and system uncertainties. This paper proposes a novel cooperative multi-agent reinforcement learning framework that integrates the MATD3 algorithm with cooperative game theory principles. The framework introduces two key innovations: (1) a Shapley value-based credit assignment mechanism within MATD3 to enhance coordination fairness, (2) a neural network-based coalition formation strategy for efficient group behavior. Experimental results demonstrate superior performance compared to baseline MATD3, achieving a stable average reward of 7.5 (vs. baseline’s 6.5) and improved linearity values around-1 (vs. baseline’s-3) The framework converges within 1,200 episodes and maintains low variance in performance metrics, indicating robust stability. Combining the subsequent algorithms advances hydraulic support control techniques and cooperative multi-agent system approaches in complex industrial environments.