基于改进堆叠自编码网络的软件质量预测方法

Software Quality Prediction Based on Improved SAE Network

随着软件在高度综合化系统中的作用日益凸显,软件质量预测成为了提高系统可靠性的有效手段。针对软件质量预测所用数据集常见的样本不足和特征冗余的问题,提出了一种基于改进堆叠自编码(SAE)网络的软件质量预测方法。首先,引入条件生成对抗网络(CGAN)来扩展数据样本。然后,使用SAE网络进行特征降维,并利用粒子群算法(PSO)优化网络结构,改善训练过程易陷入局部最优的问题。之后,结合softmax分类器完成软件质量预测。最后,通过实际机载软件项目数据仿真结果,验证了所提预测网络结构的有效性。

With the increasingly prominent role of software in highly integrated systems,software quality prediction has become an effective means to improve the system reliability.To solve the problems of insufficient samples and redundant features in the data set used for software quality prediction,an improved stacked auto encoder(SAE)network is proposed.Firstly,a conditional generative adversarial net(CGAN)is introduced to expand the data sample.Then,the SAE network is used for feature dimension reduction.To improve the problem that the training process is easy to fall into local optimum,a particle swarm optimization(PSO)algorithm is also used to optimize the network structure.Afterwards,the software quality prediction is completed with softmax classifier.Finally,the validity of the proposed predictive network structure is verified by the simulation results of actual airborne software project data.