可重构嵌入式软件缺陷优化预测仿真研究

Reconfigurable Embedded Software Defect Optimization Prediction Simulation

可重构嵌入式软件优化预测是提高软件测试效率、保证软件运行可靠性的重要手段。针对当前模型没有充分考虑软件缺陷特征历史数据的类不平衡特性,导致预测结果存在收敛速度较慢、训练误差较大的问题。提出一种基于PSO-BP算法的可重构嵌入式软件缺陷优化预测模型,充分考虑可重构嵌入式软件缺陷历史数据的类不平衡性特点,通过确定抽样度,执行欠抽样操作,缓解数据的不平衡性;同时对软件项目开发人员的能力、软件缺陷特征的数量、软件缺陷密度进行计算,并对得到的结果进行k-均值聚类;基于软件缺陷特征数据的处理结果建立具有三层网络结构的BP神经网络预测模型;运用粒子群优化算法优化BP神经网络预测模型的权值和阈值,建立可重构嵌入式软件缺陷优化预测的PSO-BP预测模型。仿真对比测试结果证明,PSO-BP预测模型与BP神经网络预测模型相比,具有更快的收敛速度,且预测结果与真实值更加接近。

This article proposes a model of optimizing and predicting reconfigurable embedded software defect based on PSO - BP algorithm. After we fully considered the class imbalance of historical data of reconfigurable em- bedded software defect, the sampling degree was determined to perform under - sampling operation, so as to relieve imbalance of data. Meanwhile, we calculated the ability of software developers, the number of software defect characteristics and the density of software defects. Then, the k - means clustering was performed on the obtained result. Moreover, BP neural network prediction model with three layers of network structure was established based on the re- sult of processing software defect feature data, and the particle swarm optimization algorithm was used to optimize weighted value and threshold value of BP neural network prediction model. Thus, PSO - BP prediction model for optimizing and predicting reconfigurable embedded software defect was built. Simulation results show that the PSO - BP prediction model has faster convergence than BP neural network prediction model, and the prediction result is close to the actual value.