基于降维技术与K-means聚类的油纸绝缘状态综合灰评估

Comprehensive grey evaluation for oil-paper insulation based on dimension reduction techniques and K-means cluster

针对变压器油纸绝缘单一特征量评估易受多种因素干扰而影响准确性等问题,提出一种基于降维技术与改进K-means聚类的变压器油纸绝缘状态综合灰评估新方法。首先,提出去极化电流谱线末端双点解析法提取3个绝缘老化新特征量。接着,基于特征量的多元评估指标阵,结合组合权重法和降维技术,提出特征量综合指标Z。最后,采用基于K-means聚类算法对Z构建"好、中、差"3等级的分级数据库并确定等级界限标准表,形成变压器油纸绝缘老化状态灰色聚类评估体系。经若干实例分析证明所提方法能全面、准确地评估绝缘状态。

Aiming at the problem that oil-paper insulation aging evaluation accuracy is vulnerable to single characteristic parameter under different circumstance, a new comprehensive gray evaluation method for transformer oil-paper insulation status based on dimension reduction techniques and K-means clustering is proposed. First, a terminal two-point analytical method of depolarization current spectrum is presented to extract three new features of insulation aging. Second, a multivariate evaluation index array is constructed and the comprehensive index Z is refined via a combined weighting method combining with main component dimension reduction techniques. Finally, a hierarchical database of ‘good, middle, and poor’ 3-level for Z is built by weighted Euclidean distance-based K-means clustering algorithm, the level limit standard table for comprehensive index is determined and the grey clustering evaluation system for transformer oil-paper insulation aging state is set up accordingly. The final numerous case analysis proves that the proposed method is comprehensive and accurate to evaluate the insulation state.