基于大语言模型的智能驾驶测试用例自动生成系统

打开文本图片集

中图分类号：U461.91 文献标志码：A DOI：10.19822/j.cnki.1671-6329.20250130

【Abstract】Traditional manual test case design has become ineffcient inaddressing theescalating complexity of modern automotive software，emerging asaprimary botteneck constraining the development cycle.To overcome this chalenge，this paper proposesan automated testcase generation methodologybasedon Large Language Models （LLMs）.The coreinovation lies inamulti-agent framework that integratespromptenginering withRetrieval-Augmented Generation （RAG）technology.Thisframework dynamicallinteractswithahierarchical，multi-sourcedomainvectorknowledgebase systemtopreciselymap high-levelnatural languagerequirementsintoexecutabletestcases.Experimentalvalidation demonstratesthatthisapproachefectivelyreplaces labor-intensivemanualdesignprocesses，significantlyenhancingthe eficiencyandconsistencyof testcasegeneration.Thispaper providesascalablesolution forintellgentqualityassurance in automotive software，significantly advancing the automation and standardization of testing workflows.

Key words： Large Language Models （LLMs），Test case generation， Retrieval-Augmented Generation（RAG），Intelligentdriving test

0引言

随着汽车产业向智能化、电动化方向加速转型，整车功能复杂度呈指数级增长，对测试验证的覆盖率与效率提出了更高的要求。（剩余18597字）