Research on Application Method of Software Reliability Model
|School||Harbin Engineering University|
|Course||Computer Software and Theory|
|Keywords||Software reliability model Model selection Model synthesis prediction BP neural network Failure data sets|
With the development of computer science,the application range of software expands rapidly, the complexity of software gets higher, and the standards of the software reliability become higher. So the researchers mainly focus on the software reliability model which is the key of the software reliability prediction.Currently, there are more than one hundred kinds of software reliability models and variations that have been published. But a simple general model dose not exist. In the practical application, there is the lack of an effective method of the software reliability model selection. On the deep research of the software reliability models and the neural network, the synthesis prediction method of the software reliability model based on the BP neural network is proposed in this paper. Firstly, some representative models as the basic models are selected from the existing models.During the selection process, we should take into account whether the candidate models promote the development of research area and whether they have the diversity.Secondly,several basic models which are suitable for the specific failure data sets are selected by the BP neural network method..Thirdly,we construct a new BP neural network and train it with the prediction data of the chosen basic models.Then,we use the trained network to predict the software failure data and compare the result with those of other models.Finally,the framework of execution procedure is presented.The experimental results indicated that synthesis prediction method of the software reliability model proposed in this paper can solve the problem to the certain extent in the model selection process. It enhances the objectivity and accuracy. In addition, the synthesis prediction models with the different characteristic models may complement mutually and improve the robustness of the reliability prediction.