Research Of Continuous-time Bayesian Network For Dependability Software Risk Prediction

As the software has been widely applied in safety-critical area, such as aviation and astronavigation, traffics and transports, nuclear power plants and medicine devices, how to enhance software dependability must be considered in the software development process. In order to advance the software dependability and demonstrate the dependability level objectively, it is important to predict software risk, to locate and remove software faults effectively. Bayesian network is a combination of probability theory and graph theory, and then a favorable tool to model and resolve uncertainty. The focus of this dissertation is to establish a framework of software risk prediction based on Bayesian network, and to study continuous-time Bayesian network models, structure and probability inference method. Thereby, it provides a systematic modeling method for software risk prediction, and it offers new technique approach and realization means for solving iterative and dynamic problem of software risk. The frameworks are rules, FTA and Bayesian network. These three methods is not only schemes of themselfs, but also complementary of each other. Experience and rules are the basis for fault tree analysis, in turn, fault tree analysis can verify the correctness of experience and rules. Meanwhile, the fault tree is the basis of Bayesian network model.By the analysis of research results in dependability software risk prediction, we have studied systemmaticly the software risk prediction.The achievments of this paper are as follows:1. This dissertation demonstrates the Closed-form reasoning process on the Bayesian network representation for the logic gates of fault tree. Based on the above closed-form formula, we can calculate the relevant continuous-time Bayesian network parameters, predict software risk and analyse software reliability.2. This dissertation proposes the framework for software risk prediction based on the continuous-time bayesian network. This framework has sufficient theoretical and practical basis, can obtain differently precise data on demand. Tt can not only help the experts, developers and managers summarize the experiences, improve software quality, but also provide some empirical data for software development.3. We use software risk prediction framework to analyse the dynamic behavior and interaction of legacy software, Combining expertise, historical data and uncertain information together, in order to improve the efficiency and credibility of modeling, predict legacy software risk and software safety assessment.It is important to predict the risk of software dependability. The methods given in this dissertation will benefit the dependability software developing in practice.