Research On Multi-component VOCs Detection Based On Sensor Arrays And Its Industrial Application Verification

Volatile organic compounds（VOCs）are important precursors of PM_2.5 and surface ozone in the atmosphere,and various kinds of VOCs have pungent smell and are toxic and harmful.Accurate and reliable detection is an important guarantee for VOCs pollution control and management.The gas sensor array has advantages in cost,detection range,volume,and other aspects,and is suitable for multi-point,low-cost,and large-scale online monitoring of VOCs pollutants.However,the VOCs emissions in industrial sites are complex and variable.Take a chemical industrial park as an example.There are large numbers of chemical enterprises concentrated in chemical industrial parks,of which the production process involves the use,storage,and transportation of a large number of raw materials,solvents,and products containing VOCs,resulting in a large amount of VOCs emissions.The application of sensor array is faced with the problems of cross-interference of various gases and low detection accuracy.In this paper,focusing on the rapid and accurate detection of multi-component VOCs,in order to solve the problems of difficult construction of sensor array,poor follow-up of traditional models,and lack of reliable training data for field applications,the sensor array configuration and feature optimization,dynamic and rapid analytical model for multi-component VOCs,sensor array and mass spectrometry coupled training method were studied.The VOCs online monitoring prototype based on the sensor array was developed,and the sensor array-online mass spectrometry VOCs monitoring system for chemical industrial parks was established,and online monitoring application verification was carried out in the typical chemical industrial park.The primary research work and conclusions are as follows:（1）In order to solve the problem of cross-interference of multi-component VOCs,a 16-channel sensor array integrating metal oxide semiconductor sensors,electrochemical sensors,and photo ionization detectors was constructed based on the principles of sensitivity,broad-spectrum response,and stability.The response spectra of the sensor array to VOCs such as toluene,ethyl acetate,methanol,ethanol,and their mixtures were analyzed.Various features（steady-state response,maximum slope of increase,response peak area,time corresponding to50%steady-state response,and time corresponding to 90%steady-state response）were extracted from the dynamic responses of the sensors.Feature selection methods based on Wilks’Λ-statistic,random forest,and other algorithms were established.The impacts of different features on the detection performance of multi-component gas were studied by combining feature selection and pattern recognition.Sensor array configuration and feature optimization were achieved.The component recognition accuracy and concentration detection RMSE of 10-fold cross-validation for low-concentration VOCs within 0.5–8 ppm are about96%and 0.53 ppm,respectively.（2）Aiming at the detection demand for VOCs components and concentrations which change continuously and rapidly,a multi-component VOCs dynamic pattern recognition model based on the fusion of gated recurrent unit（GRU）and convolutional neural network（CNN）was established.The model can automatically extract temporal features and spatial features from short-term dynamic response signals,achieving the rapid analysis of components and concentrations of VOCs mixtures.A random gas mixture sensor testing method was developed to simulate the actual emission of chemical industry parks.According to the predefined number and distribution of testing samples,VOCs mixture samples with random components and concentrations in the range of 0–60 ppm were generated to simulate the real-time measurement of the sensor array under variable components and concentrations.The GRU-CNN model can detect components and concentrations of VOCs mixtures that change continuously and rapidly for 1 minute.The component recognition accuracy and concentration detection RMSE are 91%and 2.97 ppm,respectively（relative error of 5%）.Compared with traditional models,the accuracy and follow-up of the concentration measurement of the model are significantly enhanced.（3）Aiming at the problem of lack of reliable training data in the field application of sensor arrays,a method of using online mass spectrometry to train sensor array was proposed.Firstly,a mass spectrometry signal analysis model based on multiple linear regression with L1regularization was established.In order to solve the problems of possible indefinite solutions of linear regression,a strategy of layer-by-layer search of characteristic gas database-gas calibration database-NIST mass spectrometry database was proposed to achieve accurate analysis of VOCs components and concentrations in different scenarios.A sensors’responses-pollutant composition and concentration correlation database was constructed by combining the change trend of sensors’responses to pollutants and the composition and concentration of VOCs pollutants detected by the online mass spectrometry.The sensor array was trained using this database,and the concentration detection RMSE of three kinds of VOCs（toluene,1,4-diethylbenzene,n-hexane）at a typical industrial waste gas outlet is 1.34 ppm（relative error<10%）.The accurate detection of complex VOCs emissions in the chemical industry park by the sensor array is realized,which can support the construction of a large-scale monitoring network with high precision and low cost with a small amount of online mass spectrometry and a large number of sensor arrays.（4）Aiming at the online monitoring requirements of VOCs pollutants in chemical industrial parks,a monitoring prototype based on a sensor array was developed,which can detect common VOCs such as alcohols,ketones,esters,and aromatic hydrocarbons.A sensor array-online mass spectrometry system was established,and the pollutants of typical points in the chemical industry park were monitored by the system.The monitoring results of a typical industrial waste gas outlet of the sensor array were in good agreement with those of the online mass spectrometry.Moreover,the sensor array had the advantage of high time resolution（1min）and could capture paroxysmal and instantaneous high-concentration pollution.Based on the monitoring results of the combined system,the VOCs emission characteristics of three typical enterprises in the pharmaceutical industry were analyzed.It is found that the detected VOCs pollutants are mainly produced by commonly used industrial solvents and raw materials and intermediate products in the production process.Oxygen-containing VOCs account for a relatively high proportion,such as methanol,ethyl acetate,acetone,isopropanol,etc.,which are important solvents and raw materials in the chemical industry.Then halogenated hydrocarbons follow,such as dichloromethane,ethyl bromide,chloromethane,etc.,which are mainly raw materials,solvents or intermediates or by-products in the production process.Through the study of this paper,a rapid and accurate detection method and prototype for multi-component VOCs based on a sensor array was developed,and a sensor array-online mass spectrometry system for VOCs monitoring in chemical industrial parks was constructed,which can support the construction of large-scale monitoring network in chemical industrial parks and provide guidance for pollution control and management in chemical industrial parks.