基于神经网络集成的软件故障预测及实验分析

软件系统故障预测是软件测试过程中软件可靠性研究的重点之一。利用软件系统测试过程中前期的故障相关信息进行建模，预测后期的软件故障信息，以便于后期测试和验证资源的合理分配。根据软件测试过程中已知的软件故障时间序列，利用非齐次泊松分布过程、神经网络、神经网络集成等方法对其进行建模。通过对三个实例分别建模，其预测平均相对误差G-O模型依次为3.02%、5.88%和6.58%，而神经网络集成模型为0.19%、1.88%和1.455%，实验结果表明神经网络集成模型具有更精确的预测能力。

Fault prediction and experiments analysis based on neural network ensembles

Fault prediction in software system is a key process in reliability prediction during the software testing phase. Using the forward and current correlative information of faults during the testing process to predict the fault information in afterward testing phase, it can contribute to guide the test and rationally allocate the testing resource. The fault time series known in the forward software testing process is used to model the fault predictive models by means of Non-Homogeneous Poisson Process（NHPP）, neural networks（NN）, neural network ensembles, et al. By modeling three different examples with the mean relative error of 3.02%, 5.88%and 6.58%in G-O model while 0.19%, 1.88%and 1.455%in neural network ensemble model, the result shows that neural network ensemble model has more predictive ability.