基于关联信息熵和轻量级梯度提升机的油纸绝缘特征优选策略  

Optimization strategy for oil-paper insulation features based on correlation information entropy and light gradient boosting machine

作　　者：赖汶鸿 刘庆珍[1] 鄢仁武 LAI Wenhong;LIU Qingzhen;YAN Renwu(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350108;Fujian Provincial University Engineering Research Center for Simulation Analysis and Integrated Control of Smart Grid,Fujian University of Technology,Fuzhou 350118)

机构地区：[1]福州大学电气工程与自动化学院,福州350108 [2]福建理工大学智能电网仿真分析与综合控制福建省高校工程研究中心,福州350118

出　　处：《电气技术》2024年第1期34-41,47,共9页Electrical Engineering

基　　金：国家自然科学基金项目(51807030)。

摘　　要：为了充分挖掘对变压器油纸绝缘综合诊断结果有利的老化特征量,提出一种基于关联信息熵和轻量级梯度提升机(LightGBM)的特征量优选策略。首先,基于不同老化状态的变压器介电响应实测数据,提取不同类别的时域特征量形成初始高维特征空间;其次,引入关联信息熵度量特征子集的相关性及冗余性,再利用轻量级梯度提升机评估特征的重要度,进而得到最优特征空间;最后,对比分析最优特征空间与不同对照组的诊断性能,有效验证了基于所提优选策略确定的最优特征空间的优越性。In order to fully explore the aging features,which are beneficial for comprehensive diagnosis results of transformer oil-paper insulation,a feature optimization strategy based on correlation information entropy and light gradient boosting machine(LightGBM)is proposed.Firstly,the initial high-dimensional feature space is formed with various time-domain features,which are extracted from the measured data of dielectric response of transformers in different aging states.Secondly,the correlation and redundancy of feature subsets is measured by correlation information entropy.Then the importance of features is evaluated according to LightGBM,so as to obtain the optimal feature space.Finally,the diagnostic performance of the optimal feature space is compared and analyzed against different control groups,and the superiority of the optimal feature space determined through the proposed optimization strategy is effectively verified.

关 键 词：油纸绝缘老化 综合诊断 关联信息熵 梯度提升算法 特征选择 

分 类 号：TM855[电气工程—高电压与绝缘技术]