Research Of Algorithm To Improve The Performance Of A TDLAS Gas Analyzing System

Trace gas monitoring is beneficial to atmospheric environment protection,industrial production safety assurance and production margin profit maximization.The on-line gas monitoring technology is developing rapidly,tunable diode laser absorption spectroscopy(TDLAS)gas analyzer has been widely used in real-time online detection in various industries due to its high resolution,fast response,accurate and non-contacting measurement,etc.For industry onsite application,TDLAS gas analysis system suffers multi-interferences and lower detection limit is required.The main goal of this thesis is to improve the performance of TDLAS gas analysis system.To overcome the challenges and fulfill the requirement,a TDLAS gas experimental platform is developed to verify the effectiveness of the algorithm.The gas analyzer is based on wavelength modulation spectroscopy(WMS)and the second harmonic signal is chosen as absorption spectrum to measure the concentration of different gases.First,the absorption spectra drift caused by environment changes and system degradation was studied.I propose a restoration method for spectral deformation.The deformation coefficient is obtained by the least squares support vector machine(LS-SVM).And then the horizontal expansion was restored by linear interpolation.The lateral drift was recovered by peak tracing.The least squares algorithm was used to evaluate the restoration effect,and the validity of the restoration method was verified.Second,the multi-component gas concentration analyze method was studied.In order to solve the problem of multi-component monitoring in the actual industrial production environment,single component partial least square fitting algorithm models(PLS1)and multicomponent partial least square fitting algorithm model(PLS2)were built and evaluated,together with multivariate classical least square fitting algorithm model(CLS).It showed that multiple individual PLS1 models perform the best when fulfilling the multi-component online measurement in the petrochemical process.Finally,algorithm research on weak absorption signal under strong background was studied.In this thesis,a feature-extraction-based extreme learning machine(ELM)algorithm was proposed.The absorption spectra are extracted the intrinsic feature by nonlinear iteration partial least square(NIPALS)algorithm.Then,ELM regression model is established between spectral feature vector and the corresponding concentration.Five-fold cross validation method is employed to evaluate the model prediction performance.The analysis results show that spectra pretreat with PLS feature extraction can improve the analyzer accuracy and time response,it also helps to reduce the model training time in the ELM algorithm.In this thesis,the related algorithms for system interference occurred under various circumstances is designed and the effectiveness of the algorithm is confirmed through the gas measurement experiment.These algorithms assure the TDLAS analyzer measurement accuracy in the petrochemical industry process analysis application.