融合深度学习与AM的资源数据智能校核技术  

Intelligent checking technology of resource data integrate deep learning and AM

作　　者：徐声龙 于聪 时雨欣 杨柳 胡振 XU Shenglong;YU Cong;SHI Yuxin;YANG Liu;HU Zhen(School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China;EHV Company,State Grid Hubei Electric Power Co.,Ltd.,Wuhan 430050,China)

机构地区：[1]武汉大学电气与自动化学院,湖北武汉430072 [2]国网湖北省电力有限公司超高压公司,湖北武汉430050

出　　处：《电子设计工程》2024年第19期11-15,共5页Electronic Design Engineering

基　　金：国家自然科学基金项目(71802013)。

摘　　要：在对各类资源数据进行智能校核的过程中,传统算法存在对数据量依赖性强、数据特征敏感度高的缺点。对此,文中提出了一种融合深度学习与注意力机制的改进算法。该算法针对电网人力资源数据集存在的特征维度多、原始数据特征易缺失的特点,使用KNN对数据缺失的属性进行补偿以提升数据质量,通过引入Wasserstein距离改进了对抗神经网络,以此使得少量数据集也同样能够实现较优的训练效果。采用混合注意力机制对数据特征权重加以训练,有效提升了模型的核验精度及效率。实验测试结果表明,所提算法的数据准确率均在77%以上,在对比算法中最优,具有较强的数据核验能力。In the process of intelligent verification of various resource data,traditional algorithms have the drawbacks of strong dependence on data volume and high sensitivity to data features.Therefore,this paper proposes an improved algorithm that integrates deep learning and attention mechanism.This algorithm addresses the characteristics of multiple feature dimensions and easy loss of original data features in the power grid human resources dataset.KNN is used to compensate for missing attributes to improve data quality.Wasserstein distance is introduced to improve the adversarial neural network,which enables the dataset to achieve good training results.At the same time,the use of a mixed attention mechanism to train data feature weights effectively improves the verification accuracy and efficiency of the model.The experimental test results show that the data verification accuracy of the proposed algorithm is above 77%,which is the best among the comparative algorithms and has strong data verification ability.

关 键 词：对抗神经网络 混合注意力机制 K近邻算法 数据特征 数据校核 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN915.07[自动化与计算机技术—计算机科学与技术]