Liu Yunxiao, Wang Yiming, Li Han, Wang Guangyao, Ai Jianliang
Department of Aeronautics and Astronautics, Fudan University, Shanghai, 200433, China.
Shenyang Aircraft Design and Research Institute, Shenyang, 110035, China.
Sci Rep. 2024 Sep 18;14(1):21838. doi: 10.1038/s41598-024-72808-y.
Formulating a course of action (COA) before a combat is crucial for operational command. Research in command and control (C2) artificial intelligence is currently focused on using intelligent auxiliary decision-making methods to implement COA. This paper proposes a COA planning method based on line of operation (LOO) and uses planning domain definition language (PDDL) to describe combat scenarios and COA. Following the effect-based optimization (EBO) principle, an effect evaluation model for COA was constructed, and dynamic bayesian networks (DBNs) was used to determine the reasoning and calculate the results of the effect evaluation network. To further improve the execution efficiency of the effect evaluation model in practical applications, the network was optimized through a back propagation neural network (BPNN). Relevant experiments based on the coordinated distributed air defense and anti-missile scenario were carried out using the LOO model to complete the planning of COA. A BPNN evaluation model based on the DBNs evaluation model was built. After training and fine-tuning, it achieved similar evaluation results, with a mean absolute percentage error (MAPE) of less than 0.02%. Compared with the DBNs model, the BPNN model achieved an efficiency improvement of no less than 65%, effectively reducing the consumption of computing resources. This research is the first time to realize the modeled description of COA planning, automatic evaluation, and calculation optimization of COA effects. It can support the development of decision support systems (DSS) and has the potential for practical application.
在战斗前制定行动方案（COA）对作战指挥至关重要。指挥与控制（C2）人工智能的研究目前集中于使用智能辅助决策方法来实施行动方案。本文提出一种基于作战线（LOO）的行动方案规划方法，并使用规划领域定义语言（PDDL）来描述战斗场景和行动方案。遵循基于效果的优化（EBO）原则，构建了行动方案的效果评估模型，并使用动态贝叶斯网络（DBNs）来确定推理并计算效果评估网络的结果。为进一步提高效果评估模型在实际应用中的执行效率，通过反向传播神经网络（BPNN）对该网络进行了优化。利用LOO模型基于协同分布式防空反导场景进行了相关实验，以完成行动方案的规划。构建了基于DBNs评估模型的BPNN评估模型。经过训练和微调，其取得了相似的评估结果，平均绝对百分比误差（MAPE）小于0.02%。与DBNs模型相比，BPNN模型实现了不低于65%的效率提升，有效减少了计算资源消耗。本研究首次实现了行动方案规划的建模描述、自动评估以及行动方案效果的计算优化。它能够支持决策支持系统（DSS）的发展，具有实际应用潜力。
