摘要
针对组合导航系统软件可靠性的预测问题,研究了系统累积故障数与系统运行时间的关系,提出一种基于RBF神经网络的软件可靠性预测模型,取得了理想的短期预测效果。为了获得较好的长期预测效果,采取滚动式训练、在线调整网络结构的方法对之加以改进。仿真结果表明,该模型拟合精度优于J-M模型、与G-O模型相当,预测精度高;该模型不需假设条件,泛化能力强、稳定性好,可用于实践中指导软件测试。
In order to meet the needs of accurately predicting the number of faults in program modules in integrated navigation system,the relation of the faults number and the running time was study,and a software reliability prediction mode based on RBF neural network was proposed,a predictive accuracy short dated was meet.For the sake of a predictive accuracy long ranged the neural network was improved by rolling training.The simulation results indicate that the RBF model established and the G-O model have the better quality of fit than J-M model;the RBF model shows a higher predictive accuracy than the J-M model and the G-O model;the model needs no assumption and can be used in deference applications.The model can find a better use in guiding the practical of the software test.
出处
《电子质量》
2011年第8期14-16,19,共4页
Electronics Quality
关键词
组合导航
RBF神经网络
滚动预测
软件可靠性
integrated navigation system
RBF neural network
rolling prediction
software reliability