Of Ann And Its Improved Algorithm For Atmospheric Vocs In Automatic Identification And Quantitative Analysis

The technique combining the artificial neural network (ANN) with FTIR spectroscopy was applied to the simultaneous recognizing and quantification of air VOCs. In addition, the gas dispersion in a two-dimensional plane was obtained with improved neural networks and the smooth basis function minimization (SBFM) method. Firstly, the optimum network was determined on the degree of approximation, which made the network avoid over fitting. It was shown that the single component and multi-component pollutions were identified and quantified successfully. Then, nonlinear iterative partial least squares (NIPALS) and singular value decomposition (SVD) were introduced to extract the principle components (PCs), which were used as the inputs of ANN. Results indicated that the training speed and predictive ability of ANN were all significantly improved. Comparison between them illustrated that the formal method was better. Finally, the field-measured pollution areas including VOCs were analyzed with NIPALS-ANN model and open-path FTIR spectrometer. To reconstruct the gas dispersion in a two-dimensional plane, the SBFM method was used. The area polluted can be determined and the foulest area can also be obtained. On the basis of the results, polluted source can be found as soon as possible, and measurements can be made to control diffuseness of the polluted source. This project provides a new approach for the simultaneous identification and quantification of VOCs and it also offers an effective method for real-time, automated, seriate and in situ measurements.