基于多模型变权重组合的软件可靠性仿真研究

研究提高软件可靠性预测精度问题,对软件可靠性研究已成为当前软件工程的一个研究热点,传统的单一软件可靠性模型由于使用的技术及提取的信息有限,软件可靠性预测精度不高。为提高软件可靠性预测精度,在建立多种单一软件可靠性预测模型的基础上,提出一种样本点的多模型变权重组合模型。将多种预测技术有效地聚合在一起,取长补短,在样本数据有限的情况下,不仅改善了样本内学习能力也增强了样本外的泛化能力,提高了综合预测精度。仿真验证模型无论在样本内还是样本外都较优于经过模拟退火算法优化的BP神经网络(SA-BP)及经过遗传算法优化的最小二乘支持向量机(GA-LSVM),说明变权重组合模型是一种精度更高的软件可靠性失效数据预测模型,具有较好的应用推广价值。

Software Reliability Research Based on Weight Changeable Combination Forecast Model

The study on software reliability has become one focus of software engineering.Owing to limited technique and inefficient information using,traditional sole software reliability model has unsatisfied forecast accuracy.To improve forecast accuracy,proposes one kind of weight changeable combination model based on the sample point after establishing several sole software reliability modes.On the base of limited sample data,the combine model not only improves the study ability of study sample but also buildups generalization ability by polymerizing several forecasting techniques effectively.Simulation shows that the combination model,regardless of the samples outside and inside,superior to BP net optimized by SA(SA-BP) and LSVM optimized by GA(GA-LSVM).The result proves that the combination model is one software failure model with more accuracy and has the good application promotion value.