Consistent Cooperative Control of Multi‐Support Systems Using Multi‐Agent Reinforcement Learning With Attention Mechanism

ABSTRACT The research aims to improve the coordinated management of solid filling support systems in mining to enhance efficiency, structural stability, and overall safety. Given the challenges posed by complex subterranean environments and stringent filling targets, current control methods are insufficient. This study proposes a cooperative control framework for multi‐filling support systems, modeling the consistency control problem as a multi‐agent reinforcement learning paradigm. Each filling unit acts as an agent, communicating its action state with others. A novel multi‐agent evaluation approach with an attention mechanism is introduced to facilitate path tracking and cooperative control. Considering constraints such as thrust, path deviation, and target position, simulation results show that the agents effectively collaborate to achieve the cooperative control of multi‐filling support groups in complex conditions, highlighting the significance of this approach.