基于迁移学习的嵌入式实时系统缺陷研究  

Research on the Defects of Embedded Real Time System Based on Transfer Learning

作　　者：韩邢健 曹宇[1] HAN Xingjian;CAO Yu(CEPREI,Guangzhou 511370,China)

机构地区：[1]工业和信息化部电子第五研究所,广东广州511370

出　　处：《电子产品可靠性与环境试验》2024年第1期27-33,共7页Electronic Product Reliability and Environmental Testing

摘　　要：为了减少环境变化对软件缺陷评估的影响,提出了一种基于迁移学习的嵌入式实时控制系统软件缺陷评估方法。首先,选择缺陷软件测量指标;在此基础上,利用特征聚类技术将相关的索引特征划分为同一个聚类;然后,根据两项间特征的分布相似性,找到相关特征,去除分布差异较大的特征;最后,从源项目中选择高质量的特征,构建训练数据集,通过权重调整,从源项目中选择更好的评价数据,实现对软件缺陷的准确评价。映射结果目标明确,设计方法的评价结果的错误率在8.7%以内,具有良好的评价效果。To reduce the impact of environment changes on software defect assessment,a software defect assessment method for embedded real-time control system based on transfer learning is proposed.Firstly,the defect software measurement metrics are selected.On this basis,the relevant index features are divided into the same cluster using feature clustering technology.Then,according to the distribution similarity of features between the two items,the relevant features are found,and the features with significant distribution differences are removed.Finally,high-quality features are selected from the source project to construct training dataset,and better evaluation data is selected from the source items through weight adjustment to achieve accurate evaluation of software defects.The mapping results have clear objectives,and the error rate of the evaluation results of the design method is within 8.7%,which indicates that the evaluation effect is good.

关 键 词：迁移学习 嵌入式实时控制系统 缺陷评估 测量指标 聚类技术 分布相似性 

分 类 号：TP311.56[自动化与计算机技术—计算机软件与理论]