The on-site DGA detecting and analysis system based on the Fourier transform infrared instrument

Abstract

Currently, the oil-immersed power transformer on-site detecting and analysis systems for dissolved gas analysis (DGA) were usually based on the gas chromatography, electrochemical sensor array or other sensors. But these methods had the defects such as needing standard carrier gas, regularly instrument calibration, low security, long analysis time and so on. For these reasons, an innovative method of using Fourier transform infrared (FTIR) instrument to detecting and analysis the transformer oil dissolved gas was proposed in this paper. This method has the features, such as maintenance free, no carrier gas in site, analysis time shortly (time constant below one second is possible). The detecting and analysis system was consisted by the oil sample continuous automatic collection device, oil dissolved gases separation equipment, gases analytical instruments (FTIR), data acquisition and processing module. The situation of the transformer in-site operation was simulated and a complete experimental system was set up in the laboratory. When we used the FTIR to analyze the gases on site for a long time and continuously, some key technologies (spectral baseline drift and distortion, dimension reduction and so on) needed to resolve. For the spectral baseline drift and distortion problem, the spectral baseline correction by piecewise dividing (SBCPD) baseline correction algorithm was used to preprocess the spectral data. For the problem of high-dimensional data characteristic variables selection in the spectral analysis, Tikhonov regularization algorithm was used to select spectral features variable. After the above step, the dimensions of the original spectrum was reduced from 10165 to 3~7, the computing workload was greatly reduced and the accuracy of the calculation results was improved. Finally, the sparse partial least squares algorithm was adopted to establish the quantitative analysis model of transformer dissolved gas analysis on-line monitoring system, and 7 kinds of mixed gases were analyzed. The results showed that the sparse partial least squares algorithm was superior to the traditional partial least squares algorithm. The feasibility of Fourier transform infrared spectroscopy for oil dissolved gas multi-component on-line detection and analysis system was preliminary verified by the simulation and experiment system in the laboratory.