A restoration-clustering-decomposition learning framework for aging-related failure rate estimation of distribution transformers

Distribution transformers (DTs) are critical components used in power distribution networks, and they are vulnerable to aging failures due to irreversible insulation degradation. Therefore, the accurate estimation of the aging-related failure rates (AFRs) is necessary for the reliability-centered maintenance and replacement strategies needed for ensuring service reliability and safety. Various data-intensive models have been proposed for AFR evaluation of power equipment. However, these models cannot be used for AFR evaluation of DTs due to the limitation of the available data. This paper tackles this important problem in an unconventional way by it develops a novel Restoration-Clustering-Decomposition learning framework to model the AFRs of individual DTs and improve evaluation accuracy. The proposed approach requires only the non-intrusive data that can be directly extracted from existing available databases, making it feasible to be applying to numerous DTs. First, the analysis of the degree of polymerization (DP) degradation and the Latin Hypercube sampling (LHS) technique are combined to reproduce aging failure data. Then, an optimal Entropy-weighted K-means (EW-K-means) clustering method and the classic 2-parameter Weibull model are used to evaluate the average AFRs of different DT groups through failure data analysis. Then, a DP-based decomposition function is introduced to quantify the relative aging degree of in-group individuals and to derive the probabilistic AFRs of each DT in the group. Application examples of a scrapped DT population in Chongqing Electric Power Company of China are presented and discussed in detail. The results show that the proposed learning framework has a promising capability for AFR evaluation of individual DTs and bears great practicality in the real world.