Software Reliability Model And Its Parameters Estimated

During the past four decades, many Software Reliability Models (SRMs) have been proposed and studied. They have been used for software estimation and prediction, as well as for related decision-makings in software projects such as software release time determination and optimal testing resource allocation for modular software systems. However, many proposed SRMs are based on some unrealistic or over-simplified assumptions, which to a large extent limits their applicability to real-life situations as well as their reliability estimation or prediction accuracy. These unrealistic assumptions include perfect debugging, immediate fault repair, independent software failures, etc. Effort has been put to relax these assumptions. Modified or new SRMs have been proposed in the literature.New method and analysis are explored and investigated for software reliability models and parameter estimation method. The main contents are:1. We study the parameter estimation problem for the software reliability modeling framework proposed by Trivedi, which is deal with the assumption of failure correlation. We propose a relationship function for model parameters, which appears to be essential to reduction of the number of parameters to be estimated as well as to software reliability prediction. We develop two parameter estimation methods based on different types of test data, using Maximum Likelihood Estimation (MLE) method.2. A problem existed in the general Non-Homogeneous Poisson Process model is that their assumption is too idealized for the real environment to satisfy. In this paper, we use some more rational assumptions and make some improvement based on the G-O Model, then establish a SRM which can consider imperfect debugging, time lag for fault correction and fault correlation.3. Since the parameters estimation problem is very difficult to solve, we developed an improved Genetic Annealing Evolutionary Algorithm (GAEA). First, we improve the Genetic Algorithm used in the Matlab to make the initial population more reasonable. Then, we improve the Simulated Annealing Algorithm. And its new generator and cool schedule table is optimized by the the information getting from genetic algorithm. In the end, we make the Simulated Annealing Algorithm as an independent operator embedded in the Genetic Algorithm.