Quantitative Detection of Multicomponent SF₆ Decomposition Products Based on Fourier Transform Infrared Spectroscopy Combined With CARS-ELM Algorithm

Abstract

The decomposition of SF₆ as an insulating medium in gas insulated switchgear (GIS) in the event of partial discharges leads to insulation failure inside the equipment. Therefore, it is important to achieve accurate and efficient quantitative analysis of decomposition products for effective assessment of internal insulation status. This paper proposes a quantitative calibration model based on Fourier transform infrared (FTIR) spectroscopy and feature selection algorithm combined with neural network for fast and non-destructive detection of SF₆ decomposition products concentration in GIS. Firstly, the absorption spectra of the decomposition products were obtained by FTIR spectroscopy, and then the background interference was removed according to Lambert-Beer law with 99.99% purity of SF₆ as the background interference. Subsequently, the processed experimental data were subjected to baseline correction and abnormal sample point detection before further feature selection using correlation coefficient (CC)and competitive adaptive reweighted sampling (CARS), respectively. Finally, extreme learning machine (ELM), CC-ELM and CARS-ELM quantitative calibration models were established. The results showed that the determination coefficient (R²) of CARS-ELM model reached 0.9997. It was outstanding compared with the previous two models in terms of evaluation indexes R², RMSE and MAPE performance, and the maximum relative error of prediction results for the three products under the high concentration of SF₆ interference did not exceed 1.7%. Therefore, the results prove that FTIR spectroscopy combined with CARS-ELM model is an effective method to achieve high precision quantitative detection of multi-component SF₆ decomposition products.

Topics & Concepts