非NHPP类SRGM加权组合预测精度的实验分析

EXPERIMENTAL ANALYSIS OF NON-NHPP SRGM WEIGHTED COMBINED FORECASTING ACCURACY

某些国外文献提出了NHPP(Non-homogeneous Poisson Process)类SRGM(Software reliability growth models)的统一框架,以及基于统一框架的模型组合方法,以期提高SRGM的预测能力。参考NHPP类SRGM统一框架理论,运用算术、几何、调和加权平均方法,优化非NHPP类SRGM的模型系数及其加权系数。随机选取6个失效时间间隔故障数据集,验证了某些非NHPP类(Markov类、贝叶斯)SRGM的预测能力及其加权组合的效果。实验结果表明,即使到目前为止,非NHPP类SRGM尚没有所谓的统一框架理论,但基于实数编码GA算法的加权平均方法依然可以在某种程度上提高SRGM的预测能力。

Some foreign literature put forward a unified framework of NHPP SRGM and a model combination method based on a unified framework in order to improve the SRGM prediction ability. This article referred to the NHRP SRGM unified framework theory,using the arithmetic,geometry,harmonic weighted average method to optimize non-NHPP SRGM model coefficients and their weighting coefficients. We randomly selected six failure interval datasets to verify the effectiveness of some non-NHPP class（Markov class,Bayesian） SRGM prediction ability and its weighted combination.The experiment result show even without a unified scheme between Non NHPP（Markov and Bayesian） SRGM by now,the combination methods improve the prediction ability of SRGM and combined SRGM to some extent.