Further Research On The Unascertained Software Reliability Model And Search For New Solution

With the increasingly wide range and importance of software applications, the software quality requirements for the people are getting higher and higher. Software reliability models as the core and key, can be used at different stages of the software life cycle. Quantitative estimating and predicting of software reliability is of vital importance at software distribution of resources, decision-making software market. Although software reliability model has been developed to the stage from the research phase of the project, but in the face of the software development process itself and its increasingly complex situation, these models still show some limitations, the most prominent problem is the inconsistency of the models.Blind number theory as a tool for the expression of unascertained information and computing tools, which is the expansion of unascertained mathematics, is more general in description of some unascertained problems and it expands the information based on uncertainty design optimization space, so there is great potential and bright prospects for the future. Blind number theory used in this paper to improve the model based on the unascertained theory, and then establishes the software reliability model based on blind number. The process of modeling is introduced in detail expression by the testing case of the Armored Force Engineering Institute as an example, and then, combined NTDS, SYS series failure data with other classical model, unascertained reliability model by the statistic results of reliability parameters, as well as short-term forecasts and the model evaluation criteria (such as Prequential Likelihood, Model Bias, Bias Tendency, and so on). Through computing and the simulation tests, confirmed the new model which has the higher forecast accuracy than other models, and it has a more extensive adaptation, achieves a certain degree of optimization results.In addition, this paper searchs for the new method for modeling, and brings forward model, which has the adaptive genetic operators in genetic programming, for the software reliability. Genetic Programming has no influence from the various data sets' special nature and thus there is no necessary to assume that randomness as the basic characteristics of the data. Furthermore, this method doesn't require to understand the inherence processes for failures, but to create models based on the given data for a "true" process during the specific modeling course. Thereby effectively descripts of the failure mechanism of software, by adding adaptive genetic operators algorithm can effectively prevent "premature" phenomenon, and quickly finds the global optimum. This paper based on the failure of the five classic data sets as an example, introduces the modeling processes. And then, contraposes the reliability parameters, as well as short- term forecasts and capacity model evaluation criteria (such as Prequential Likelihood, Model Bias, Bias Tendency, etc.) and other model parameters compared by calculation, simulation tests, compared with the traditional model. It confirmes the new model has a high prediction accuracy, is a good model of consistency, to a certain extent, improves the problem of low accuracy, poor adaptability of the existing models, and is of positive significance in riching the thinking and system of the software reliability modeling.