Estimation Method for Nonuniform State of MOIP in Traction Transformer via Curve Reconstruction of Frequency Domain Spectrum

Multilayer oil-immersed insulation paper (MOIP) is widely employed in different kinds of traction transformers, of which the insulating state decides the safe operation. Generally, the insulating state of MOIP is nonuniform distribution, bringing great challenges to the accurate estimation. Based on the idea of curve reconstruction of frequency domain spectrum (FDS), this article proposed an estimation method for the nonuniform state of MOIP using double extreme learning machine (ELM) networks and an improved cuckoo search (ICS) algorithm. First, the MOIP was divided into different equivalent parts that have the same insulating state. Then, the hybrid equivalent circuit of the MOIP was established and solved by the counting-ELM network and the proposed ICS algorithm. Then, the FDS of MOIP can be reconstructed. Consequently, the FDS of each equivalent part can be obtained. Finally, the insulating states of each equivalent part can be estimated by the classifying-ELM network. The results show that the nonuniform state of MOIP can be estimated effectively by the proposed method. Meanwhile, the hybrid equivalent circuit can reflect the polarization characteristics of MOIP, and the proposed ICS algorithm has better performance than the traditional cuckoo search (CS) algorithm.