估测组合导航系统软件缺陷的一种神经网络方法

NEURAL NETWORK BASED PREDICTION OF SOFTWARE FAULTS IN INTEGRATED NAVIGATION SYSTEM

在早期的软件开发寿命周期里 ,软件的质量模型包含确定有缺陷的模块和预测错误数 .在大规模组合导航软件系统中 ,质量控制的主要任务就是要精确地估测程序模块中存在的缺陷数 .本文给出了用神经网络技术估测程序中的缺陷数的方法 ,预测缺陷数一般用单隐层网络就可以 ,因为多隐层网络将相应占用更多的训练时间 .本文还介绍了一种新的静态可靠性模型 ,根据软件复杂度 ,估测了质量及质量拟合度 ,并且与传统回归技术进行了比较 ,仅用单隐层神经网络模型就比传统的统计模型预测质量好 .当应用于一批大型组合导航软件系统的数据时 ,神经网络技术表明 ,它的确有更好的质量拟合度和估测能力 ,它给软件工程提供一个好的管理工具 .

Presents the neural network based prediction of software faults in integrated navigation system including the prediction of number of faults by a single layer of network only to reduce training time,and a new static reliability model suitable for prediction of quality and quality fit depending upon complexity of software,and the comparison with the traditional regression technique which shows that the prediction with a single hidden layer neural network achieves better quality than that with the traditional statistic model,and concludes that the neural network technique has better quality fit and prediction capability when it is applied to the data for a large_scale integrated navigation system and it provides a good managemant tool for software engineering.