Concentration Inversion Of Multi-component Volatile Organic Compounds Based On Deep Neural Network

In recent years,deep neural networks have shown advantages in automatically acquiring data features and abstracting expression data relationships.Nonlinear regression prediction based on deep neural networks has been widely used in engineering practice and experimental research.In view of the fact that shallow artificial neural networks rely on prior knowledge to extract features manually,and shallow network structures limit the ability of neural networks to learn complex nonlinear relationships,this paper focuses on multi-component VOCs concentration inversion based on deep neural networks.It is applied in the online monitoring of VOCs based on Fourier transform infrared spectroscopy(FTIR).The main work of this paper is as follows:The FTIR measurement system was established under laboratory conditions using a project-made Fourier transform infrared spectrometer to measure three typical VOCs of benzene,toluene and o-xylene.The experimental results show that high-intensity infrared spectrum information can be obtained by increasing the optical path of the gas pool by multiple reflections.Eight VOCs such as benzene,toluene,1,3-butadiene,ethylbenzene,styrene,o-xylene,m-xylene and p-xylene were selected from the US Environmental Protection Agency(EPA)database.The absorbance spectrum in the wavelength range of 12?m was used as the original data,and the concentration of eight kinds of VOCs was inversely analyzed by deep neural network multivariate nonlinear correction method.The simulation results show that the deep neural network regression prediction model can realize the multi-component VOCs concentration inversion,and the predicted root mean square error of the inversion is 0.0027 ppm,which is compared with the predecessors using nonlinear partial least squares.The accuracy of fitting,artificial neural network and other methods has been significantly improved.The root mean square error of each VOC gas does not exceed 0.005 ppm,and the root mean square error of each sample does not exceed 0.006 ppm,which proves that the deep neural network prediction model has good nonlinear fitting ability.And good stability.When the training sample is insufficient(typical value: less than 500),the deep neural network cannot fully learn,the network error is larger,and the accuracy is lower than that of the single hidden layer artificial neural network,but as the number of training samples increases,the deep neural network The accuracy is continuously improved.When the number of training samples is sufficient,the deep neural network has stronger nonlinear relation learning ability than the shallow artificial neural network,and the prediction accuracy is higher and the model is more stable.Experiments show that deep neural networks can fully learn data features without manual extraction of features,and at the same time perform concentration inversion of multi-component VOCs and achieve high precision.The dimensionality reduction of spectral matrix before training greatly reduces the complexity of the algorithm.Degree,effectively improve the efficiency of inversion.Studies have shown that deep neural networks have a good application prospect in the inversion of multi-component VOCs based on infrared spectroscopy.