基于最优等级数的多维特征量开关柜健康状态评价方法

Evaluation for Switchgear Health States Based on Multi-dimensional Feature with Optimal Rank

为更加客观准确地依据带电检测数据对开关柜的健康状态进行评价,提出了一种基于最优等级数的多维特征量开关柜健康状态评价方法。首先,基于局部放电数据的波动特性,引入波动程度指标,构建局部放电多维特征数据库,充分挖掘局部放电数据特征;然后通过K均值聚类评价算法中不同簇的误差平方和确定最优簇的个数K即评价等级数,解决评价等级数的主观设定问题,并基于该最优等级数对开关柜局部放电健康状态进行聚类分析;最后采用t分布随机近邻嵌入降维算法实现了高维度数据聚类的可视化。通过现场带电检测数据验证了算法的可行性,与传统方法对比,该文所提方法有效地将开关柜健康状态分类的准确率提高了10.9%,为开关柜运维检修中健康状态的评估提供了一定的理论依据。

In order to evaluate the health state of switchgear with the live detection data more objectively and accurately, we propose an evaluation method for health states based on the multi-dimensional features and optimal rank. Firstly, based on the fluctuating characteristics of partial discharge data, the index of fluctuating degree is introduced. Meanwhile, a comprehensive partial discharge multi-dimensional feature database is constructed to fully exploit the characteristics of partial discharge data. Then, the number of optimal rank K in the cluster algorithm is determined by the sum of squared errors, and the K is the optimal rank. This can solve the problem of setting the number of rank subjectively in the cluster algorithm. Also, the cluster algorithm is applied to classify the health state of switchgear. Finally, the t-distributed stochastic neighbor embedding algorithm is used to reduce dimension to visualize the cluster results two-dimensionally to realize the visualization of high-dimensional data. The feasibility of the algorithm is verified by the live detection data, which can effectively improve the classification accuracy of switchgear health states by 10.9%. Thus it can provide a certain theoretical basis for the evaluation of switchgear.