考虑多因素的大规模设备分区检修决策

Decision of Large-scale Equipment Partition Maintenance Considering Multi-factor

针对大规模设备检修决策问题,提出了基于谱聚类算法的多因素设备分区检修模型。首先考虑设备健康指数、设备故障水平、设备价值、社会影响、设备地位等多方面因素建立了设备检修指标的样本空间;然后通过Euclidean距离构造设备相似度矩阵;以设备相似度矩阵的Laplace矩阵相对特征值差自动确定设备分区检修数目,以Laplace矩阵特征向量为依据,确定设备分区检修方案;最后根据以上检修思路,以江西某市级电网为例进行验证。算例结果表明,所提出的谱聚类算法将Laplace矩阵特征向量分别映射到2维和3维向量空间,当相对特征值差处于峰值时,设备检修分区数目为3,故可将设备划分为3个检修分区,从而达到大规模设备分区检修的效果;此外,非规格化Laplace矩阵和规格化Laplace矩阵的分区相关系数分别为0.842 3和0.842 5,说明规格化Laplace矩阵的检修分区效果更好,验证了该算法的有效性。

A partition maintenance model of equipment is established for large-scale equipment maintenance based on spectral clustering. First we built a sample space based on the health index, historical levels index,equipment value index, social impact, and equipment status index; then we built a similarity matrix using the Euclidean distance, determined the number of partition maintenance of equipment by the Laplace relative eigengaps of similarity matrix, and divided parti- tion maintenance of equipment based on Laplace matrix. Finally, we took the power grid of Jiangxi as an example to verify the partition maintenance model. The example results show that the spectral clustering can map the first two or three eigenvectors of Laplace matrix to the two-dimensional or three-dimensional eigenvector space, the number of parti- tion maintenance of equipment is three when relative eigengaps maximum, so we can divide partition maintenance of equipment for three to realize large-scale partition overhaul. Moreover, the partition correlation coefficients of non-normalized Laplace matrix and normalized Laplace matrix are 0.842 3 and 0.842 5. It is verified that normalized La- place matrix partition is more effective, and the proposed algorithm is effective and feasible.