基于NHPP模型组合的失效时间预测

摘要

为了提高软件可靠性模型的预测精度，本文提出了一种基于NHPP可靠性模型的组合预测模型。该模型以几个经典的NHPP模型为基预测模型，通过贝叶斯方法推导出模型在某一次失效中占的权重值与其在上一次失效中各模型占的权重值的关系式，再运用最大熵原理获得预测误差的分布，迭代得出各个预测模型的权重。考虑到由时效性引起的很久前的信息导致了对当前信息判断的误差，所以将迭代次数设为固定常数而不再迭代到初始时间，提高了模型对软件可靠性的预测精度。实验的结果也表明，本文所研究的模型在预测的精度上好于单一模型，对比传统的组合模型以及不考虑时效性的模型也都有更好的表现。 In order to improve the prediction accuracy of software reliability models, this paper proposed a combination prediction model based on NHPP software reliability models, first using the Bayesian Approach to deduce the relation between the weight of every single model in one software failure and the previous failure; second, obtaining the distribution of prediction error by Maximum Entro-py Principle, then getting the weight of every single model by iteration. Considering that the failure occurred long time before should not affect the information we obtained in recent times, we set the iteration times as a constant, which improved the prediction accuracy. The results of experiment show that the combined model this paper proposed performs better than single model and other traditional combined models on prediction accuracy.