Author(s)

Liu Hongjun, Guo Jinran, Zhang Jiyue, Yang Han

Corresponding Author

Liu Hongjun

Abstract

In order to solve the problem of rapid diagnosis of fault in value-added service mode of special transformer users, based on the integration platform of measurement automation, an on-line fault diagnosis method for power supply equipment of special transformer users is designed. Firstly, the time series autoregressive model of state parameters of special transformer equipment is established, and the time series is quantized as the input value of the system by using the self-organizing neural network. Learning samples of least squares support vector machines are established based on process input in sliding time window. Then the deviation between the regression calculation results and the measured values of the eigenvectors of the model is set as the observation value. Gauss mixture model is used to fit the distribution of multidimensional observation values to build up the background model of the system. The fault index is calculated by the matching degree between the new individual observation value and the background model, thus real time diagnosis of equipment faults can be achieved. Experimental results show that this method can predict fault online quickly and accurately.

Keywords

On-line fault diagnosis, Self-organizing neural network, Least squares support vector machines, Fault index.