基于AI增强的技术路线图构建方法研究

摘要/Abstract

摘要：

技术路线图对于我国新质生产力发展具有重要战略意义。为提升技术路线图构建的效率、准确度和智能化水平,研究提出基于AI增强的技术路线图构建框架:首先,针对技术-产品功能-市场三维度的主题演化分析,分别基于技术评价指标体系、大语言模型信息抽取、大语言模型生成式摘要+聚类及技术-关系-技术(Technology-Relation-Technology,TRT)语义分析等方法识别各维度内主题,并分析演化过程。其次,针对维度间关联构建,基于大语言模型提取中间产品功能层的技术点及市场应用,通过主题词匹配构建维度间关联。再次,针对技术预测,以未来数据搭建的知识库为基础,基于检索增强生成(Retrieval-Augmented Generation,RAG)技术增强大语言模型,通过与其交互,预测未来应用场景,并面向场景预测技术发展。最后,以无人机目标识别与跟踪领域为例进行实证分析,验证了方法的科学性与可行性。

关键词: 技术路线图, 人工智能, 大语言模型, 检索增强生成, 无人机

Abstract:

The technology roadmap holds significant strategic importance for the development of new quality productive forces in China. However, the existing methods for constructing technology roadmap still need to make trade-offs between ease of use and accuracy, and the intelligence level also needs to be improved. Moreover, the accuracy of mining information in technology roadmap based on network data source still needs to be optimized. As artificial intelligence enters a new stage, how to simultaneously enhance the efficiency, accuracy, and intelligence level of technology roadmap construction method based on AI technology is an urgent issue to be solved today.Therefore, given the existing limitations in current research, this study proposes an AI-enhanced framework for technology roadmap construction. Firstly, conduct theme evolution analysis for the three dimensions of "technology" - "product function" - "market" respectively. In the technology dimension, based on the evaluation index system of frontier technologies, the data set is subject to theme clustering through the screening of patents and academic papers, and the theme evolution of the technology dimension is analyzed in combination with the time dimension. In the product function dimension, the text classification of Large Language Model(LLM) is used to screen product news related to the field, and based on this, information extraction of product categories and functions is carried out, and the product theme evolution is analyzed in combination with the time dimension. In the market dimension, for two types of data sources, namely news, scientific and technological reports, the methods of "generative summary by LLM + clustering" and "TRT semantic analysis" are respectively adopted, and the market theme evolution is analyzed in combination with the time dimension. Secondly, the themes among various dimensions are associated through keywords matching. The technical points and market applications in the product function sentences are extracted based on LLM, and are respectively matched with the theme words of the technology dimension and the market dimension to establish the associations among various dimensions. Finally, based on the Retrieval-Augmented Generation (RAG) technology and the interactive dialogue with LLM, the framework implements systematic prediction of technological development.Utilizing the knowledge database built with futuristic data, through the multi-round interaction with the RAG-enhanced LLM, the future application scenarios are comprehensively predicted, and based on these scenarios, the technological development trends in the future are scientifically forecasted, completing the entire technology roadmap construction process. An empirical analysis is conducted in the field of UAV target identification and tracking, constructing a technology roadmap for this field. Three realized technical routes are identified, including radar detection, computer vision, and laser detection. The analysis reveals three characteristics of future technological development in the field of UAV target identification and tracking, namely "systematization," "intelligence," and "modularity". This study provides strong support for the development of the UAV target identification and tracking field and validates the scientificity and feasibility of the proposed framework, which makes the drawing of the technology roadmap more efficient, accurate, and convenient, and enhances the intelligence level of the node construction and prediction process of the technology roadmap.

Key words: technology roadmap, artificial intelligence, large language models, retrieval-augmented generation, UAV

中图分类号:

F204

苗红, 王浩桐, 吴菲菲. 基于AI增强的技术路线图构建方法研究[J]. 科学学研究, 2026, 44(6): 1192-1205.

MIAO Hong, WANG Hao-tong, WU Fei-fei. Research on the AI-enhanced method for technology roadmap construction[J]. Studies in Science of Science, 2026, 44(6): 1192-1205.

图/表 12

图1 研究思路图

Figure 1 Research framework

图2 基于大语言模型的产品类型及功能等信息提取流程图及效果评估方法

Figure 2 Process and effect evaluation of information extraction for product types, functions, etc. based on LLM

图3 产品新闻信息提取任务提示词工程

Figure 3 Prompt engineering for the task of product news information extraction

表1 基于大语言模型的产品新闻信息提取任务效果评估

Table 1 Evaluation of the effectiveness of product news information extraction based on LLM

任务描述	效果评估
判断领域相关新闻文本分类任务	精确率93.57%,召回率97.58%,F1分数95.52%
产品类别、名称、供应商信息挖掘任务	前50项数据中有44项三元组信息全部正确
产品功能信息挖掘任务	前50项数据中有43项提取全面准确

图4 产品功能识别结果

Figure 4 Identification results of product function themes

图5 产品功能句技术点、市场应用挖掘任务提示词工程

Figure 5 Prompt engineering for mining technical points and market applications in product function sentences

图6 技术层与产品功能层关联构建

Figure 6 Construction of the association between technical layer and product function layer

图7 产品功能层与市场层关联构建

Figure 7 Construction of the association between product function layer and market layer

表2 构建知识库收录的未来数据信息

Table 2 The futuristic data information included in the knowledge base

数据源	出版方	出版物类型	数量
政府网站	中华人民共和国中央人民政府	政府新闻	1
政府相关官方组织	北京市人民政府	短期规划	1
	美国空军部(Department of the Air Force)	中长期规划	1
	美国国防部(U.S. Department of Defense)	中长期规划	1
政府智库	美国国会研究服务部(The Congressional Research Service)	智库报告	1
民间机构及组织	麻省理工学院科技评论(MIT Technology Review)	专业人士评论	1
民间机构及组织	无人争锋(远望智库旗下无人装备领域前沿资讯平台)	公众号文章	2

图8 基于RAG增强大语言模型的技术预测响应过程示意图

Figure 8 Diagram of the technical prediction process based on RAG-enhanced LLM

图9 无人机目标识别与跟踪领域技术路线图

Figure 9 Technology roadmap in UAV target identification and tracking

表3 识别的现有前沿技术节点与基于未来场景的预测技术节点对照

Table 3 Comparison between the identified existing frontier technical nodes and the predicted technical nodes based on future scenarios

识别的现有前沿技术节点	基于未来场景驱动的预测技术节点
雷达传感器探测技术	传感器信息融合技术
信号滤波处理
激光探测与测距技术
视觉系统与图像处理技术
目标识别与跟踪算法	人工智能算法
无线通信技术	通信感知一体化技术
波束成形技术
跟踪路径规划技术
任务规划技术
飞行控制技术	超视距自主飞行技术
系统稳定性控制	抗干扰与自适应技术
/	实时数据处理技术
/	载荷模块化
/	隐形技术

参考文献 29

[1]	Phaal R, Farrukh C J P, Probert D R. Technology roadmapping—A planning framework for evolution and revolution[J]. Technological Forecasting and Social Change, 2004, 71(1-2): 5-26. DOI URL
[2]	Kim J, Park G, Woo M, et al. How big data has changed technology roadmapping: A review on data-driven roadmapping[J]. IEEE Access, 2025, 13:8297-8309. DOI URL
[3]	Pu X, Gao M, Wan X. Summarization is (almost) dead[J]. arXiv preprint arXiv:2309.09558, 2023.
[4]	Phaal R, Farrukh C, Probert D R. T-Plan: The Fast-start to Technology Roadmapping? Planning Your Route to Success[M]. Institute for Manufacturing, University of Cambridge, 2011.
[5]	Phaal R, Farrukh C, Probert D R. Strategic roadmapping: A workshop-based approach for identifying and exploring strategic issues and opportunities[J]. Engineering Management Journal, 2007, 19(1): 3-12.
[6]	Li X, Zhou Y, Xue L, et al. Integrating bibliometrics and roadmapping methods: A case of dye-sensitized solar cell technology-based industry in China[J]. Technological Forecasting and Social Change, 2015, 97: 205-222. DOI URL
[7]	Wei T, Jiang T, Feng D, et al. Exploring the evolution of core technologies in agricultural machinery: A patent-based semantic mining analysis[J]. Electronics, 2023, 12(20): 4277. DOI URL
[8]	Kim J, Geum Y. How to develop data-driven technology roadmaps: The integration of topic modeling and link prediction[J]. Technological Forecasting and Social Change, 2021, 171: 120972. DOI URL
[9]	Miao H, Wang Y, Li X, et al. Integrating technology-relationship-technology semantic analysis and technology roadmapping method: A case of elderly smart wear technology[J]. IEEE Transactions on Engineering Management, 2020, 69(1): 262-278. DOI URL
[10]	Geum Y, Lee H J, Lee Y, et al. Development of data-driven technology roadmap considering dependency: An ARM-based technology roadmapping[J]. Technological Forecasting and Social Change, 2015, 91: 264-279. DOI URL
[11]	Feng L, Niu Y, Wang J. Development of morphology analysis-based technology roadmap considering layer expansion paths: Application of TRIZ and text mining[J]. Applied Sciences, 2020, 10(23): 8498. DOI URL
[12]	Kim S, Jang H, Yoon B. Developing a data-driven technology roadmapping method using generative adversarial network (GAN)[J]. Computers in Industry, 2023, 145: 103835. DOI URL
[13]	Wang Z, Pang Y, Lin Y, et al. Adaptable and Reliable text Classification using Large Language Models[C]. 2024 IEEE International Conference on Data Mining Worshops (ICDMW). IEEE, 2024: 67-74.
[14]	马冰琦, 周盈海, 王梓宇, 等. 一种基于大语言模型的威胁情报信息抽取方法[J]. 网络空间安全科学学报, 2024, 2(2):36-46. DOI
	Ma B Q, Zhou Y H, Wang Z Y, et al. A threat intelligence information extraction method based on large language models[J]. Journal of Cyberspace Security Science, 2024, 2(2): 36-46.
[15]	NVIDIA. What is retrieval-augmented generation, aka RAG? [EB/OL]. https://blogs.nvidia.com/blog/what-is-retrieval-augmented-generation/. 2025-01-31/2025-02-14.
[16]	Kresevic S, Giuffrè M, Ajcevic M, et al. Optimization of hepatological clinical guidelines interpretation by large language models: A retrieval augmented generation-based framework[J]. NPJ Digital Medicine, 2024, 7(1): 102. DOI PMID
[17]	中国政府网. 国家中长期科学和技术发展规划纲要[EB/OL]. https://www.gov.cn/jrzg/2006-02/09/content_183787.htm. 2023-11-23.
	China Government Network. National medium - and long-term science and technology development plan outline[EB/OL]. https://www.gov.cn/jrzg/2006-02/09/content_183787.htm. 2023-11-23.
[18]	苗红, 连佳欣, 李伟伟, 等. 基于数据挖掘的前沿技术识别方法与实证研究[J]. 系统工程与电子技术, 2024, 46(9):3082-3092. DOI
	Miao H, Lian J X, Li W W, et al. Empirical research and method of frontier technology identification based on data mining[J]. Systems Engineering and Electronics, 2024, 46(9): 3082-3092.
[19]	苗红, 王浩桐, 李伟伟, 等. 面向应用场景的前沿技术识别方法[J]. 情报杂志, 2025, 44(1):95-103.
	Miao H, Wang H T, Li W W, et al. Frontier technology identification method for application scenarios[J]. Journal of Intelligence, 2025, 44(1): 95-103.
[20]	王东清, 芦飞, 张炳会, 等. 大语言模型中提示词工程综述[J]. 计算机系统应用, 2025, 34(1):1-10.
	Wang D, Lu F, Zhang B, et al. Survey on prompt engineering in large language model[J]. Computer Systems Applications, 2025, 34(1):1-10.
[21]	朱琳. 专利数据驱动的产品技术创新机会识别与评价方法研究[D]. 宁波大学, 2019.
	Zhu L. Research on the method of product technological innovation opportunity identification and evaluation driven by patent data[D]. Ningbo University, 2019.
[22]	汪玲. 基于SAO结构挖掘的智能语音技术演进及其特征研究[D]. 大连理工大学, 2021.
	Wang L. Research on the evolution and characteristics of intelligent voice technology based on SAO structure mining[D]. Dalian University of Technology, 2021.
[23]	Choi S, Kim H, Yoon J, et al. An SAO-based text-mining approach for technology roadmapping using patent information[J]. R&D Management, 2013, 43(1): 52-74. DOI URL
[24]	周杰. 科技报告资源的构成及产生机理研究[J]. 情报学报, 2013, 32(5):466-471.
	Zhou J. Research on the composition and generation mechanism of scientific and technical report resources[J]. Journal of the China Society for Scientific and Technical Information, 2013, 32(5): 466-471.
[25]	贺德方. 科技报告的内涵、作用与管理机制[J]. 情报学报, 2014, 33(8):788-792.
	He D F. The connotation, role and management mechanism of scientific and technical reports[J]. Journal of the China Society for Scientific and Technical Information, 2014, 33(8): 788-792.
[26]	Kim J, Park Y. Supporting Technology Foresight for Disruptive Innovation: Keyword-based Visual Analysis For Futuristic Data[C]. Proceedings of the 2014 Asia Pacific Industrial Engineering & Management Systems Conference, 2014.
[27]	Jang W, Kim J, Park Y. Comparing the patent data and futuristic data based on keyword and network analysis[J]. International Journal of Arts & Sciences, 2015, 8(6): 573.
[28]	Kim J, Park Y, Lee Y. A visual scanning of potential disruptive signals for technology roadmapping: Investigating keyword cluster, intensity, and relationship in futuristic data[J]. Technology Analysis & Strategic Management, 2016, 28(10): 1225-1246.
[29]	Jang W, Seol H. Identifying the Emerging Safety Technologies Using Futuristic Data: Topic Modeling and Diversity Approach[C]. International Conference on Management Science and Engineering Management. Cham: Springer International Publishing, 2018: 251-263.