基于机器学习的软故障定位技术研究
Abstract
Withthescaleofsoftwaresystemshasbeenexpandingandthestructurehasbecome
complexincreasingly,inthiscase,inordertoimprovethedebuggingefficiencyofsoftware
programs,howtoaccuratelydetectprogramfaultsremainsachallengingtask.Toaddressthe
issueofincompletecoverageoftestcasebranchesinexistingdatasets,inthispaper,improved
zebraoptimizationalgorithmisproposesedfortestcasegeneration.Inaddition,inviewofthe
lowefficiencyofexistingmethodsinsoftwarefaultlocationandtheinabilitytoeffectively
connectthecontext,convolutionalneuralnetworksisappliedtothefieldofsoftwarefault
location,takingtheprogramsourcecodeastheresearchobjecttocarryoutresearchon
softwarefaultlocationtechnology.Thespecificresearchcontentisconductedasfollows:
AtestcasegenerationmodelZOA-GAisestablished.Tosolvetheproblemthattest
casescannotfullycoverbranchpaths,atestcasegenerationmodelZOA-GAisproposed
whichintegratesGeneticAlgorithm(GA)andZebraOptimizationAlgorithm(ZOA).Firstly,
thismodelcombinesstaticanalysisandprograminstrumentationtoobtainpathinformation.
Secondly,thefitnessfunctionisconstructedthemethodoflayerproximityandthetechnique
ofoverlappingpaths.Finally,tosolvetheproblemthatZOAiseasilyfallingintolocaloptima,
GAisintroducedintoZOA.TheexperimentalresultsshowthatcomparedtostandardZOA,
GA,andparticleswarmoptimizationalgorithms,theZOA-GAmodelrequiresfewerrounds
togeneratetestcaseswhenthepopulationsizeisdifferentcomparedtootheralgorithms.
AsoftwarefaultlocalizationmodelFL-ResNetisEstablished.Insoftwarefaultlocation
technology,mostcurrentcannotbelinkedtothecontex