Stud On Inversion Method Of Pollutant Gas Concentration Based On Genetic Algorithm Combined With Limit Learning Machine

Atmospheric pollution is very serious.The development of gas detection technology to achieve real-time,efficient and accurate monitoring of harmful gas content is an urgent problem to be solved in the current field of polluted gas research.Fourier transform infrared spectroscopy(FTIR)technology has become an ideal means widely used in the field of gas pollutant monitoring because of its significant advantages such as no sampling,wide detection range,high sensitivity,high detection accuracy and multi-component gas continuous on-line monitoring.Focusing on the common problems in gas concentration inversion based on FTIR technology,such as large dimension and high collinearity of spectral data,serious aliasing of spectral lines and complex background noise,this paper takes the gas concentration inversion algorithm model as the main line,and carries out the following research work:1)In view of the insufficient fitting accuracy of the traditional linear regression method to the nonlinear relationship and the common shortcomings of the traditional neural network method,such as slow training speed,unstable solution and weak generalization ability,the combination of genetic algorithm and extreme learning machine(ELM)is proposed for gas concentration inversion.Based on the principle research of PLS,BP-ANN,elm and GA,the design of GA-ELM combined algorithm is carried out,and the theoretical feasibility and advantages of the algorithm are verified.2)Based on Lambert Beer’s law,the mixed absorbance models of 9 common low concentration volatile toxic gases are established.After appropriate preprocessing of the data,the parameters of PLS,elm and GA-ELM algorithms are optimized according to the characteristics of mixed gas infrared spectrum data,and finally the concentration inversion is carried out.The results show that GA-ELM can well realize the spectral unmixing and concentration inversion of multicomponent polluted gases,and the mean square error of each component gas in the GA-ELM(5.02×10-12)model is significantly lower than that of the ELM(9.58×10-12)and PLS(19.60×10-12)models,which proves that the GA-ELM algorithm has good nonlinear fitting ability and anti-interference.3)Starting from the whole process of FTIR gas concentration inversion,the concentration inversion experimental system is designed and built by using selfdeveloped FTIR equipment.The standard concentration gas is obtained by using highprecision gas distribution system,and the detection of sulfur hexafluoride gas with complex background noise under open light path is completed.Based on the experimental site conditions,the spectral restoration and preprocessing of the data are carried out,and then the algorithm optimization and model building of PLS,ELM,GABP and GA-ELM algorithms are carried out to complete the inversion of gas concentration.The experimental results show that GA-ELM can well realize the concentration inversion of small sample gas under complex noise background,The mean square error,standard deviation and optimal relative error are 63.04×10-12,6.33×10-12,0.4%,respectively.By comprehensively comparing the running time,mean square error and corresponding standard deviation of each model,it can be concluded that GA-ELM algorithm has higher prediction accuracy,running efficiency,better stability and generalization performance.This paper verifies the feasibility of FTIR technology in gas concentration inversion in principle and technology through experiments.The research results show that genetic algorithm combined with extreme learning machine method is an effective gas concentration inversion method based on FTIR technology,and has a good application prospect in the field of gas concentration inversion based on FTIR technology.