Application Research On Software Reliability Modeling By Genetic Programming

As the comprehensive application and increasing importance of software, people's requirement to software quality becomes higher and higher. Software reliability models, which are considered as the kernel and critical for the reliability appraisal, can be used for different phrases in software life-cycle and estimate or predict software reliability behaviors quantitatively. A good reliability model can estimate and predict software reliability action accurately, which plays a significant role on software resource allocation and software market decision–making. Software reliability models have already stepped engineering stage from the conceptual phase, however, they still appears inherence shortcomings facing the increasingly complicated software and development courses. Among which, the incongruous problem in model application is the prominent one.On the basis of no influence on various data sets' particularity, consequently, Genetic Programming(GP for short) does not need to presume randomicity as the basic characteristic of the data.Also,it will not concern whether the data is linearity or homogeneity non-linearity (that is the non-linearity system which can transform to the linearity system).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, which can describe the software failure mechanisms more effectually and predict for the next failure times more exactly. This paper firstly adopts GP algorithm to hunting model, which can possibly reflect system behaviors, in the function spaces compoundly constituted by the authorized function operators.Meanwhile,we have proved that GP can obtain the best solution for failure behavior's variation rules from the convergence character of itself, which provides a new thought for software reliability modeling. Moreover, this paper makes use of Particle Swarm Optimization (PSO for short) to adjust the parameters, and finally puts forward the method for software reliability predicting based on GP.This paper takes three classical examples for failure data. The process of GP modeling is introduced in detail. Comparing with other models, the statistic results of reliability parameters, short-predicting capability and the model appraisal criteria (such as Prequential Likelihood, Model Bias, Bias Tendency, and so on) are given. Through calculations and simulations, it proves that the accuracy of new model is higher and the dependency for sample data is much lower. Due to this approach can break away from the probabilistic assumptions about the variety of failure rate, the inconsistency of application shows obviously comparing with other traditional models, which can enrich the mentality and system of software reliability modeling to some extent. This research is hopeful to reform the problems of low-accuracy and poor-adaptability for the models in existence, which will take on aspiring merit for enriching the ideology and system of software reliability modeling.