基于时序数据的软件可靠性模型组合新方法

New combination approach of software reliability model based on time-series data

基于时序数据的软件可靠性模型受到越来越多的关注,然而单一模型在精确度和通用性上都存在不足,鉴于此,提出一种新的软件可靠性模型组合的方法,该方法将反向传播(BP)神将网络模型和支持向量机回归(SVMR)模型进行组合,通过遗传算法(GA)和滑动窗口机制构造可靠性模型输入,使用粒子群(PSO)算法选择单一模型的最优参数,并使用BP神经网络确定两个模型的权重值建立组合模型,来预测下一阶段的软件失效数据。最后进行了仿真实验并做了对比分析,结果表明该方法较单一模型具有更高的精确度和较好的通用性。

Software reliability models based on time-series data attract more and more attention, however, single model has shortcomings in both accuracy and versatility. This paper proposed a new combination approach of software reliability models, which combines Back Propagation（ BP） neural network model and the Support Vector Machine Regression（ SVMR）model. This approach established a model through a combination of Genetic Algorithm（ GA） and the sliding window mechanism, used Particle Swarm Optimization（ PSO） algorithm to choose the optimal parameters of single model, and used the BP neural network to determine weights of the two models to predict the software failure data of next phase. Finally,experiments were carried out. The comparative analysis of the experimental results reveals that this proposed method has higher accuracy and better versatility than single model.