基于FABP的软件可靠性增长模型

Software Reliability Growth Model Based on FABP

由于单一神经网络建立的软件可靠性预测模型的预测精度不高且适用性差,用高级神经网络建立的软件可靠性预测模型的网络结构过于复杂。为了提高软件可靠性预测模型的适用性和在保证预测精度的情况下降低神经网络的结构,提出了利用软件缺陷数据,在BP神经网络训练过程中利用萤火虫算法(Firefly algorithm,FA)对BP神经网络的权值和阈值进行寻优,同时采用多次预测结果取均值的方式来减小BP神经网络预测的波动性的方法来建立基于FABP的软件可靠性预测模型。利用3组软件缺陷数据,以误差比均值和误差平方和作为衡量标准进行模型预测。仿真结果表明,用FABP建立的软件可靠性预测模型具有较高的预测精度和适用性并且具有相对简单的网络结构。

Prediction accuracy of software reliability prediction model（SRPM） based on a single neural network is not high and this SRPM has low adaptability, and network structure of SRPM based on advanced neural network is too complex. In order to improve the adaptability of SRPM and reduce the neural network structure in the case of high prediction accuracy, a new prediction model is proposed. The weights and thresholds of the BP neural network are optimized by Firefly Algorithm （FA） in the training process of BP neural network by software defect data. At the same time, in order to reduce the fluctuation of prediction by BP network, averaging method is used to deal with pre- dicted results. Based on those, software reliability prediction model is established by FABP. According to 3 groups of software defected data, and the mean value of error and sum of squared errors are taken as measurement to compare prediction performance. Simulation results show that the SRPM based on FABP which has a relatively simple network structure can improve the prediction accuracy and adaptability.