Search

The chemical industrial park is an important source of volatile organic compounds (VOCs) emissions which is difficult to monitor due to its diversity and low concentration. In this work, a hybrid sensor array consisting of six gas sensors (four semiconductor sensors, one electrochemical sensor, and one photo ionization detector) was constructed to measure VOCs. Three most frequently detected VOCs in a Chinese fine chemical industrial park, namely toluene, dichloromethane, ethyl acetate, and their mixtures, were tested all below 5 ppm (even below 1 ppm in many cases). Support vector machine (SVM) and back-propagation artificial neural network (BP-ANN) based on different input features, including steady-state responses (SS), the first four components preprocessed by principal component analysis (PCA) and linear discriminant analysis (LDA), were trained and evaluated. The 10-fold cross validation results indicate that SS-SVM and SS-ANN are two best models with accuracy and $\text{R}^{{2}}$ of about 94% and 0.95 for the validation set, respectively. Compared with LDA, PCA is more suitable for dimension reduction in this study, since the performance of models with the first 4 PCs are closer to that of models with SS. The results demonstrate that the proposed hybrid sensor array system with SVM and ANN models is able to identify and quantify low-concentration VOCs mixtures. This method is promising to be used for low-cost and online monitoring of VOCs emissions in chemical industrial parks. [ABSTRACT FROM AUTHOR]