The Research On Software Error Data And Software Reliability Measure Model

With the advent of information age, computer technology hasmadeleapsandbounds, andsoftwarehasbeenwidelyusedinvariousfield-s of social life, which has become the core element of the modernization.Software system attends in all aspects of social daily life, which has affect-ed the development of the society to some extent. Therefore, the demandfor software quality is increasingly high and to the system failure toleranceability is getting worse. Reliability becomes the key indicator to measuresoftware quality and those that have low reliability cannot come out eventhough they have good functions. Different from hardware reliability, soft-warereliabilityanalysisbothfromthetheoreticalresearchandpracticalap-plication is at the early stage, which also contributes to be a hot researchfiled in the recent years. The software reliability model is the earliest andthe most fruitful in reliability study and still is the most active direction. Itcan quantitatively describe the errors behaviour of software during the testand use process, and can be a more accurate way to predict the reliabilityof the assessment system. Guiding the resources effectively in the entireprocess, the method can assure the software release in time and meanwhileachieve the given reliability level. Therefore, how to use the software er-ror data to construct mathematical models becomes the worthy direction offurther mining in the software reliability research.In this thesis, based on the premise of analysing the existing softwarereliability models, we focus on constructing novel models and improvingthe existing model by new method applying based on software error data inordertomakethemodelbemoreefficient. Maininnovationsareasfollows.Firstly, a software reliability growth model is established on the basisof compounded-decreased-rate model, drawing on classic software models based on the rate of removing errors. The new model takes the situationof multiple factors into account and describes a more detailed process ofthe software error by promoting the rate of removing errors, which is moreconsistent with the reality and improves the feasibility. After using publicavailable data sets to validate the effectiveness of the proposed model, theresults indicate that the new model has an advantageous performance ofboth fitting and predicting.Secondly, fullytakingintoaccountthesoftwareerrorprocessandchar-acteristic of the error data, a software reliability growth model based onchange point is proposed by integrating a change point into the modelthrough the unified theory. The Change Point Analysis (CPA) was adoptedto assess the credibility of the change point while enhancing the ability todetermine change point such as c-chart. Thus, the proposed method hasbetter operability in practice, and effectively promotes the development ofsoftwarereliabilitymodelsbasedonthechangepoint. Moreover, thevalid-ity of the model has also been approved by the application of internationalpublic software error databases.Thirdly, onthebasisofthesoftwarereliabilitymodelbasedonstochas-tic process theory and the unified theory about NHPP model, combing thecounting process and pure birth process, a software reliability model basedon pure birth process has been established in this thesis. And then a newlyimproved reliability model based on pure birth process that has advanta-geous performance in reality has been constructed through the hypothesisof better software error process. The feasibility of the proposed model wasproved through theoretical analysis and model comparison.