Decision reliability analysis framework for intelligent unmanned system-of-systems using the cooperative multi-agent deep deterministic policy gradient

The decision reliability is the capability of the intelligent unmanned system-of-systems (IUSoS) to make and execute decisions under a disturbed environment. However, the inherent uncertainty of individual autonomy may render decisions unreliable, leading to task failures and resource wastage. To address this issue, we propose a novel decision reliability analysis framework for the IUSoS using the cooperative multi-agent deep deterministic policy gradient (C-MADDPG). Firstly, considering individual autonomy and functional diversity, a multilayer operation network based on graph coloring theory is established to provide a foundation for decision analysis. Secondly, a distributed decision model based on the enhanced actor-critic (AC) architecture is presented to maintain the individual stable decision-making ability. Thirdly, considering the impact of individual decisions on reliability, an operation-loop-oriented decision reliability assessment model is proposed to quantify decision reliability. Then, the decision reliability is optimized by the C-MADDPG algorithm, which updates the enhanced AC network with the cooperative reward. Finally, a case study on a multi-agent collaborative scenario demonstrates the feasibility of the methodology. The proposed method receives the most rewards compared with variant algorithms and consistently outperforms them in task completion degree, providing a technical approach for improving system-of-systems reliability under distributed architectures and demonstrating that the cooperative reward mechanism balances individual autonomy with system reliability.