Evaluation Of Software Reliability Based On Cloud Theory And Gray Model

In the field of evaluation of software reliability, dominating evaluation models are probability statistics based models which generally ignore the fuzziness of reliability concept. In the field of prediction of software reliability, machine learning based methods need a large number of samples and a long training time. Traditional gray models(GM) need only a few samples, but those models can’t handle the data which contains abnormal value. To solve those problems, cloud focus theory based evaluation model and Random Sample Consensus(RANSAC)algorithm based new gray model are deeply researched in this dissertation. The main research work of this dissertation can be summarized as follows:1.There search background of prediction and evaluation of software’s reliability, the research status of relative fields have been summarized. In addition, some relative theories have been overviewed, such as cloud theory and gray theory, which provide a theoretical basis for the following research works.2.One weight determination method of evaluation index of software reliability is proposed. The new method combines the objective entropy method and subjective factor discriminant table method. The weights of evaluation index of software reliability are obtained via normalization operation which is based on the weights produced by the two mentioned methods.3.The cloud focus theory is introduced into the evaluation of software reliability. The cloud focus theory based evaluation model of software reliability is proposed. Taking a series of quantitative index as the input, providing a bridge to quantitative data and qualitative concept, this novel model can produce an intuitionistic reliability level. The effectiveness of this model has been proved by simulation experiment.4.The RANSAC algorithm is introduced into the prediction of software reliability. The RANSAC algorithm based GM prediction model is proposed. The RANSAC algorithm is used for detecting and excluding abnormal data and preferential policy is used for selecting the condition of fixed solution in this new model which improves the defects in traditional GM prediction model. The new model is tested on five datasets, telecom network fault record data, naval tactical data system fault record data and so on. The result of simulation experiments shows that improved GM prediction model is better than traditional GM prediction model.