Software Reliability Forecasting Method Based On Decomposition And Reconstruction Of Series

In the field of software reliability, reliability prediction method based on the failure datamodeling mainly make research on the relationship between software reliability and software failureby making use of stochastic process and statistical analysis method or by machine learning and timeseries analysis method. Both the classical and data-driven software reliability models can not afford toobtain satisfactory results in most data sets due to the software failure data in the actual collectionprocess which interferences with human factors and other objective factors, making the finalstatistical results inevitably carry some noise.In this paper, on the basis of fully analyzing software failure data, studying on software reliabilityprediction analysis method on account of failure data sequence decomposition and reconstruction isput forward in order to reduce the effects of software failure data sequence noise and improve thereliability of the model fitting and forecast precision and generality.This article has the followingthree innovations:Firstly, the software reliability prediction model based on the residual sequence. According to thecharacteristics of the ARIMA model with GEP algorithm, the model which is more extensive andconsistent with actual situation is presented in further study of ARIMA reliability prediction modelwith GEP.Secondly, the software reliability prediction model based on empirical mode decomposition.Using EMD methods to extract different characteristics of data sequences of software failurebehaviors and combined with support vector regression and grey prediction theory forecastpreprocessing data, the final prediction results can be gained by refactoring the prediction results.Thirdly, the software reliability prediction model based on singular spectrum decomposition.SSA decomposition approach are exploited to decompose software failure data sequence into severalindependent subsequences which are respectively for predictive modeling using the theory of greyprediction and ARIMA.At last, we can conclude that appropriate to the various models of the scene by discussing ofexperimental results of the three models.