Application Of Support Vector Machine In Quantitative Analysis Of Multi-Component Mixed Gas

Along with China's economy continues to improve,the level of car ownership increased year by year,environmental pollution problems have become increasingly serious.So car emissions testing work has become very important.Application of vehicle exhaust sensor to complete emissions testing work,it`s achieved using NDIR method to measure mixing harmful gas of emissions at the same time.When we measure the multi-component gases simultaneously,multiple gas absorption lines will overlapped,it means there is a certain absorption interference.In addition,other gases in the environment(e.g.,H2 O and O2)will affect the measured gases.This disturbances belong to nonlinear problems,if not be corrected,it will seriously affect the sensitivity and accuracy of the measurements.This article takes multiple gas density sensor which is based on the dispersion of light infrared method(NDIR)as the object of study,elaborates in detail the domestic and foreign multi-component gas concentration detection technology development,and the technology of cross-interference correction.Then it introduces the principle of the detection system,including NDIR detection principle,infrared absorption principle and theoretical analysis,in particular the principles of nonlinear modeling and analysis techniques,in order to reduce the influence of cross-inference.Finally,this article introduces Support Vector Machine to model,combined with grid search method and particle swarm optimization algorithm for modeling parameters optimization,the model is applied to the multi-component gas quantitative analysis.Then this article uses BP neural network which is optimized by genetic algorithm to analysis the multi-component gases.Further,compared with the method of support vector machines,as we can see from the experiment results: the SVM algorithm is superior in dealing with cross-interference.