基于WKPCA与IEDO-XGBoost的变压器故障诊断方法研究  

Transformer fault diagnosis method based on weighted kernel principal component analysis with improved exponential distribution optimization XGBoost

作　　者：张容槟 徐耀松[1] 牛元平 ZHANG Rongbin;XU Yaosong;NIU Yuanping(Faculty of Electrical and Control Engineering,Liaoning Technical University,Huludao 125105,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,辽宁葫芦岛125105

出　　处：《电工电能新技术》2024年第10期24-42,共19页Advanced Technology of Electrical Engineering and Energy

基　　金：国家自然科学基金项目(51974151);辽宁省教育厅重点实验室项目(LJZS003);辽宁省教育厅辽宁省高等学校基本科研项目(LJ2017QL012);辽宁省教育厅科技项目(LJ2019QL015)。

摘　　要：针对变压器故障的特点,将加权核主成分分析技术与IEDO-XGBoost相结合,提出了一种新的变压器故障诊断模型。该方法主要将溶解气体分析技术与无编码比值法相结合,获取变压器的故障特征,利用WKPCA对其进行降维处理,并将归一化处理后的故障样本数据作为IEDO-XGBoost模型的输入,输出变压器故障诊断类型及其诊断准确率。选取20维变压器故障特征数据进行WKPCA降维处理,加快了模型的收敛速度;采用自适应正余弦策略和高斯变异策略对指数分布优化器算法进行改进,并用10个典型测试函数对改进后的指数分布优化算法性能进行了测试,结果表明改进后的指数分布优化算法具有更快的收敛速度和全局搜索能力。然后,利用改进的指数分布算法来确定XGBoost模型中的多个最优参数。仿真结果表明,该模型的诊断准确率为91.82%,分别比EDO-XGBoost、NGO-XGBoost、GJO-XGBoost、GWO-XGBoost和WOA-XGBoost故障诊断模型高2.73%、3.64%、5.46%、8.18%和10.91%,验证了本文所提方法能够有效提高变压器故障诊断性能。Aiming at the characteristics of transformer faults,a new transformer fault diagnosis model is proposed by combining the Weighted Kernel Principal Component Analysis(WKPCA)technique with IEDO-XGBoost.The method mainly combines the dissolved gas analysis technique with the non-coded ratio method to obtain the fault characteristics of the transformer,use WKPCA to reduce its dimension,and use the processed normalized fault sample data as the input of the IEDO-XGBoost model to output the transformer fault diagnosis type and its diagnostic accuracy.The 20-dimensional transformer fault feature data are selected for WKPCA dimension reduction processing,which accelerates the convergence speed of the model;the exponential distribution optimizer algorithm is improved by using the adaptive sine-cosine strategy and Gaussian variance strategy,and the performance of the improved exponential distribution optimization algorithm is tested by using 10 typical test functions.The results show that the improved exponential distribution optimization algorithm has faster convergence speed and global search ability.Then,the improved exponential distribution algorithm is used to determine multiple optimal parameters in the XGBoost model.Simulation results show that the diagnostic accuracy of the model is 91.82%,which is 2.73%,3.64%,5.46%,8.18%and 10.91%higher than that of EDO-XGBoost,NGO-XGBoost,GJO-XGBoost,GWO-XGBoost and WOA-XGBoost fault diagnosis models,respectively,which verifies that the proposed method can effectively improve transformer fault diagnosis performance.

关 键 词：变压器 加权核主成分分析 故障诊断 溶解气体分析 指数分布优化算法 极端梯度提升 

分 类 号：TM41[电气工程—电器]