2025 International Conference on Equipment Intelligent Operation and Maintenance (ICEIOM2025)

3 / 2025-03-17 11:01:19

Research on Cross-Domain Combat Methods Based on PPO-DRL Algorithm

PPO-DRL,Task allocation,Motion planning,Reconnaissance strikes,Decision-Making

Special Sessions > SSL 12 Advanced Techniques for Machinery Prognostics and Health Management Based on Data-driven Methods

In the exploration and research of intelligent warfare, cross-domain combats have emerged as an effective means for system disruption and are considered the preferred mode of operation to achieve combat objectives. This study focuses on the reconnaissance-strike task scenario in cross-domain combat and addresses the issues of task prediction allocation and motion planning management in complex and uncertain environments. The research first elaborates on the complexity of task decision-making and the uncertainty of the battlefield environment from the perspectives of battlefield environment model parameterization and typical swarm reconnaissance-strike tasks. Subsequently, the study designs a generalized state space, reward function, action space, and policy network, while incorporating diverse rewards closely related to the reconnaissance-strike task. The action strategy output adopts a subject-verb-object structure to more accurately represent complex operations. The policy network employs an encoder-temporal aggregation-attention mechanism-decoder architecture to effectively integrate feature information. The study utilizes a deep reinforcement learning method based on proximal policy optimization (PPO-DRL) to solve the problem. Finally, simulation experiments validate the feasibility and effectiveness of decision-making for cross-domain combat entities performing reconnaissance-strike tasks under complex and uncertain conditions, demonstrating the intelligence of task prediction allocation and motion planning management.