Crossover response processing technology of photoacoustic spectroscopy signal of SF6 decomposition components under partial discharge

Abstract

Overlap exists in the infrared absorption band of some SF6 decomposition components under partial discharges (PD), so there would be serious crossover response and precision decline when photoacoustic technology is used to detect these components. Thus, we combined principal components analysis (PCA) with radial basis function(RBF) neural network to construct a PCA-RBF neural network which suppresses crossover responses. The newly constructed neural network is applied in processing the output signal array of photoacoustic detection to solve the problem of precision decline for using tradition RBF neural network in the case of serious correlated input space, and to achieve accurate detection of concentrations of gas components in a gas mixture of SO2, CO2, and CF4. The results show that the PCA-RBF neural network is effective in eliminating correlation between samples, and it raises the detection precision of neural networks to concentration of various components in gas mixture(the average relative error of this method drop to less than 3%).The research provides an effective method of data processing for using photoacoustic spectroscopy in detection of SF6 decomposition components under partial discharge.