基于模块化推理的态势认知思维链技术研究

doi:10.16182/j.issn1004731x.joss.25-0984

摘要/Abstract

摘要：

针对传统仿真系统中态势理解智能化不足等问题，构建了态势视觉问答数据集，提出了模块化推理框架。构建了态势认知思维链SACoT，在零样本条件下通过专家提示引导模型进行任务分解与多模态信息融合，生成推理链以增强语义认知和可解释性，提供了一种可扩展、低计算成本的解决方案。实验结果表明：SACoT优化了任务分配，使模型聚焦于问题相关的图像细节，减轻了多步推理引起的思维链碎片化，缓解了模型在长文本推理过程中的遗忘问题，验证了模块化分析在态势推理中的可行性，为AI应用于战场态势仿真融合与作战辅助决策提供了新思路和技术途径。

关键词: 基础模型, 思维链, 态势推理, 视觉问答, 零样本学习

Abstract:

To address issues such as insufficient intelligence of situational understanding in traditional simulation systems, a situational visual question answering dataset was constructed, and a modular reasoning framework was proposed. The SACoT was built, which, under a zero-shot setting, employed expert prompts to guide the model in task decomposition and multimodal information fusion, generating reasoning chains to enhance semantic cognition and interpretability and offering a scalable solution with low computation cost. Experimental results indicate that SACoT improves task allocation, enables models to focus on query-relevant image details, mitigates the fragmentation of chain-of-thought induced by multi-step reasoning, and reduces long-form text forgetting during the reasoning process. This validates the feasibility of modular analysis for situational reasoning, offering a new approach and pathway for applying AI to battlefield situation simulation fusion and combat decision support.

Key words: foundation model, chain-of-thought, situation reasoning, visual question answering, zero-shot learning

中图分类号:

TP391.9

姬鸿远,卿杜政 . 基于模块化推理的态势认知思维链技术研究[J]. 系统仿真学报, 2026, 38(2): 278-293.

Ji Hongyuan,Qing Duzheng . Research on Chain-of-thought Technology for Situational Awareness Based on Modular Reasoning[J]. Journal of System Simulation, 2026, 38(2): 278-293.

图/表 19

图1

图2

图3

图4

图5

图6

图7

图8

表1

图9

图10

表2

图11

图12

表3

图13

图14

图15

表4

参考文献 32

[1]	Endsley M R. Toward a Theory of Situation Awareness in Dynamic Systems[M]//Salas E. Situational Awareness. London: Routledge, 2011: 9-42.
[2]	王昊奋, 易侃, 吴蔚, 等. 多模态态势感知的知识表示、表示学习和知识推理[J]. 指挥信息系统与技术, 2022, 13(3): 1-11.
	Wang Haofen, Yi Kan, Wu Wei, et al. Knowledge Representation, Representation Learning and Knowledge Reasoning for Multi-modal Situational Awareness[J]. Command Information System and Technology, 2022, 13(3): 1-11.
[3]	张大永, 杨镜宇, 吴曦. 兵棋推演空中任务智能预测方法研究[J]. 系统仿真学报, 2023, 35(1): 212-220.
	Zhang Dayong, Yang Jingyu, Wu Xi. Research on Intelligent Prediction Method of Wargaming Air Mission[J]. Journal of System Simulation, 2023, 35(1): 212-220.
[4]	陈晓婧, 朱德政. 基于人工智能的战场态势感知和作战辅助决策[C]//第九届中国指挥控制大会论文集. 北京: 中国指挥与控制学会, 2021: 580-584.
	Chen Xiaojing, Zhu Dezheng. Battlefieldsituation Sensing and Operational Auxiliary Decision-making Based on Artificial Intelligence[C]//Proceedings of the 9th China Conference on Command and Control. Beijing: Chinese Institute of Command and Control, 2021: 580-584.
[5]	张杰, 王刚, 姚小强, 等. 基于防空反导的战术级通用态势图分析研究[J]. 火力与指挥控制, 2020, 45(2): 37-42.
	Zhang Jie, Wang Gang, Yao Xiaoqiang, et al. Analysis and Research of Tactical Level General Stuation Map Based on Air Defense and Antimissile[J]. Fire Control & Command Control, 2020, 45(2): 37-42.
[6]	周洁静, 蒋婷婷. 通用空战场综合态势图的研究[J]. 信息化研究, 2020, 46(4): 1-6, 11.
	Zhou Jiejing, Jiang Tingting. Research on Common Air Battlefield Comprehensive Situation Picture[J]. Informatization Research, 2020, 46(4): 1-6, 11.
[7]	乔殿峰, 梁彦, 马超雄, 等. 多域作战下的群目标意图识别与预测[J]. 系统工程与电子技术, 2022, 44(11): 3403-3412.
	Qiao Dianfeng, Liang Yan, Ma Chaoxiong, et al. Recognition and Prediction of Group Target Intention in Multi-domain Operations[J]. Systems Engineering and Electronics, 2022, 44(11): 3403-3412.
[8]	冷志成, 薛凤桐, 张灏龙, 等. 基于智能仿真推演的导弹防御指挥控制系统研究[J]. 软件工程, 2021, 24(8): 39-43.
	Leng Zhicheng, Xue Fengtong, Zhang Haolong, et al. Research on Missile Defense Command and Control System Based on Intelligent Simulation Deduction[J]. Software Engineering, 2021, 24(8): 39-43.
[9]	Antol S, Agrawal A, Lu Jiasen, et al. VQA: Visual Question Answering[C]//2015 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2015: 2425-2433.
[10]	Lu Pan, Mishra S, Xia T, et al. Learn to Explain: Multimodal Reasoning Via Thought Chains for Science Question Answering[C]//Proceedings of the 36th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2022: 2507-2521.
[11]	Vaswani A, Shazeer N, Parmar N, et al. Attention Is All You Need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.
[12]	Devlin J, Chang Mingwei, Lee K, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding[C]// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg: ACL, 2019: 4171-4186.
[13]	Brown T B, Mann B, Ryder N, et al. Language Models Are Few-shot Learners[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2020: 1877-1901.
[14]	Radford Alec, Wook Kim Jong, Hallacy Chris, et al. Learning Transferable Visual Models from Natural Language Supervision[C]//Proceedings of the 38th International Conference on Machine Learning. Chia Laguna Resort: PMLR, 2021: 8748-8763.
[15]	Yao S, Zhao J, Yu D, et al. ReAct: Synergizing Reasoning and Acting in Language Models[C]//The Eleventh International Conference on Learning Representations. 2022: 1-33.
[16]	Wei J, Wang Xuezhi, Schuurmans D, et al. Chain-of-thought Prompting Elicits Reasoning in Large Language Models[C]//Proceedings of the 36th International Conference on Neural Information Processing Systems, 2022: 24824-24837.
[17]	Zhang Zhuosheng, Zhang A, Li Mu, et al. Multimodal Chain-of-thought Reasoning in Language Models[EB/OL]. (2023-02-02) [2024-12-01]. .
[18]	Houlsby Neil, Giurgiu Andrei, Jastrzebski Stanislaw, et al. Parameter-efficient Transfer Learning for NLP[C]//Proceedings of the 36th International Conference on Machine Learning. Chia Laguna Resort: PMLR, 2019: 2790-2799.
[19]	Hu E, Shen Yelong, Wallis P, et al. LoRA: Low-rank Adaptation of Large Language Models[C]//ICLR 2022 Conference. New York: ICLR, 2022: 1-13.
[20]	Li Xiang, Liang P. Prefix-tuning: Optimizing Continuous Prompts for Generation[EB/OL]. (2021-01-01) [2025-09-28]. .
[21]	Kojima Takeshi, Gu Shixiang, Reid Machel, et al. Large Language Models Are Zero-shot Reasoners[C]//Proceedings of the 36th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2022: 22199-22213.
[22]	Wan Xingchen, Sun Ruoxi, Dai Hanjun, et al. Better Zero-shot Reasoning with Self-adaptive Prompting[EB/OL]. (2023-05-23) [2024-12-20]. .
[23]	Dong Qingxiu, Li Lei, Dai Damai, et al. A Survey for in-context Learning[EB/OL]. (2022-12-31) [2024-09-14]. .
[24]	Gao Timin, Chen Peixian, Zhang Mengdan, et al. Cantor: Inspiring Multimodal Chain-of-thought of MLLM[C]//Proceedings of the 32nd ACM International Conference on Multimedia. New York: ACM, 2024: 9096-9105.
[25]	Mitra C, Huang B, Darrell T, et al. Compositional Chain-of-thought Prompting for Large Multimodal Models[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 14420-14431.
[26]	Shao Hao, Qian Shengju, Xiao Han, et al. Visual CoT: Advancing Multi-modal Language Models with a Comprehensive Dataset and Benchmark for Chain-of-thought Reasoning[C]//Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 8612-8642.
[27]	Liu Aixin, Feng Bei, Xue Bing, et al. DeepSeek-V3 Technical Report[EB/OL]. (2024-12-27) [2025-04-01]. .
[28]	Comanici Gheorghe, Bieber Eric, Schaekermann Mike, et al. Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities[EB/OL]. (2025-10-16) [2025-08-30]. .
[29]	Wang Peng, Bai Shuai, Tan Sinan, et al. Qwen2-VL: Enhancing Vision-language Model's Perception of the World at Any Resolution[EB/OL]. (2024-10-03) [2025-04-03]. .
[30]	Zhu Jinguo, Wang Weiyun, Chen Zhe, et al. InternVL3: Exploring Advanced Training and Test-time Recipes for Open-source Multimodal Models[EB/OL]. (2025-04-19) [2025-09-01]. .
[31]	Zheng Ge, Yang Bin, Tang Jiajin, et al. DDCoT: Duty-distinct Chain-Of-thought Prompting for Multimodal Reasoning in Language Models[C]//Proceedings of the 37th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2023: 5168-5191.
[32]	Chen Qiguang, Qin Libo, Zhang Jin, et al. M³CoT: A Novel Benchmark for Multi-domain Multi-step Multi-modal Chain-of-thought[EB/OL]. (2024-05-26) [2024-11-15]. .

推理要素	主要内容示例	主要信源	推理要素	主要内容示例	主要信源
环境条件	天气，气候，地形，设施	Deepstate	时空态势	速度，高度，方位，位移	SCSPI
行动策略	进攻，佯攻，伴攻，防御	Deepstate	攻势类型	钳形，楔形，迂回，游击	Deepstate
装备选择	坦克，导弹，无人机，等	ISW	航迹意图	贸易，监视，挑衅，演习	SCSPI
地区职能	后勤，指控，火炮，战略	ISW

分数	细则
1	没有给出答案或者生成的答案与正确答案所指的事物、选项或情况完全无关
2	生成的答案包括多个猜测，有正确答案或部分正确答案，但也可能提到错误答案
3	生成的答案只包含部分正确答案或仅包含正确答案的一个方面，不包含错误答案
4	生成的答案与正确答案相同

方法	平均单样本推理时间/s
Baseline	0.9
CoT	1.3
MMCoT	8.5
CCoT	9.2
VisCoT	4.4
SACoT	7.8

流程消融	召回率	准确率
SACoT(无描述流程)	81.04	67.11
SACoT(无决策流程)	81.74	65.67
SACoT(无集成流程)	81.67	63.22
SACoT	81.90	67.78