Early Prediction Model Of Software Reliability On Component-base Software

Software reliability is one of the most important indicators of software quality. Software reliability model evolved from analyzing data flow and failure data, into modeling component-based software architecture and component failure behavior, which is called as Component Base Reliability Model (CBRM), as the developing of software engineering and software development mode, and the CBRM based on Markov chain has aroused general interest. In order to make the software reliability model capable of guiding software production and providing reliable software quality assurance, more and more studies focused on the early prediction model of software reliability. This thesis proposes a brand new modeling program base on this kind of CBRMs.In the early period of software development, base on the formalization of UML use case diagram, sequence diagram, we can obtain the Markov chain model of software architecture. When Markov chain is a type of stochastic finite automaton whose output has nothing to do with its input, this thesis gives the algorithm that can transfer and simplify the Markov chain into stochastic finite automaton considering the characteristic of software architecture model. Then, the concept of metric entropy is introduced as a measure of software reliability. According to the definition of metric entropy, this thesis deduces the algorithm to calculate the metric entropy through the transition probability matrix. By investigating the definition of metric entropy, it clarifies its meaning in the respect of software reliability measurement, which is come of the information system reliability measurement, and then discusses the relationship between metric entropy and the traditional software reliability model. Base on the software architecture model, this thesis uses the improved neural network to model the component failure behavior at a disadvantage of lacking information about components. And it comes out the whole early prediction model of software reliability on component-base software.This article introduces the previous research and concepts involved first, and then give the method to modeling the software structure and component failure behavior combined with a meteorological data collection system as an example. The processing of model the software architecture is the core, which involved a number of algorithms, the relationship between Markov chain and the stochastic finite automaton, deducing the functional relationship between metric entropy and the traditional software reliability model.