Research On The Combination And Optimization Of Software Reliability Models Based On ANN And LSSVM

With the rapid development of computer technology, the software reliability is more and more important in the computer system. How to predict the software reliability and prevent the occurrence of catastrophic accident has become the important research topic in the field of software engineering. In the process of the actual software reliability prediction, different software systems often need to choose different software reliability model. ANN has strong adaptive ability and learning ability, and it has good performance in software reliability prediction. However, it can't jump out of the limitations of computational intelligence. LSSVM can effectively describe the software reliability from a statistical perspective, but its preferences depend on expert experiences, and prediction accuracy of system is difficult to improve.In view of the above problems, the ANN based software reliability model and the LSSVM based software reliability model is combined in this paper, and an analytic selection based simulated annealing algorithm(ASSA) to optimize the combination model is proposed, the major work is listed as follows:Firstly, the software reliability theory, the SVM theory and common software reliability model is reviewed, in order to provide theoretical basis to solve more complex prediction problems.Secondly, with ANN and LSSVM as the foundation, the fuzzy fitness variable weight method is used, which combines two kinds of software reliability model and constructs the new combination forecast model(CO-ANN-LSSVM).Thirdly, an analytic selection based simulated annealing algorithm is proposed to trained the weights of RBF neural network in the network, and the artificial parameters of LSSVM is optimized. Finally, the feasibility and effectiveness of the model is verified by Matlab simulation experiment. Experimental results shows that the constructed model CO-ANN-LSSVM has a good predict effect on software system.