基于静态删减算法的神经网络在软件可靠性预测中的应用

摘要

Following a study of typical software reliability prediction models,a new software reliability prediction model based on neural network using SPA (Silent Pruning Algorithm) is proposed.The model is different from most software reliability prediction models based on neural networks in that the topological structures of the networks are determined by the algorithms with prior knowledge or trial and error.Simulation with two suites of classical software defect information show that SPA helps improve the accuracy of neural network-based software reliability prediction models and the relationship between the topological structures of neural networks and prediction accuracy is discussed in comparison with traditional neural network-based software reliability prediction models.%与传统的基于人工神经网络的软件可靠性预测模型结构普遍由预先经验或者通过不断尝试的方法确定不同,在分析一般的软件可靠性模型的基础上,提出将一种基于静态删减算法的神经网络模型应用到软件可靠性预测中.通过利用2组经典的软件失效数据进行仿真,并与传统的神经网络可靠性预测模型对比分析,研究了神经网络拓扑结构和预测精度的关系,结果表明,基于静态删减算法的运用能够提高神经网络模型对软件可靠性的预测精度.