Artificial neural network for the quantitative analysis of air toxic VOCs

A 23-6-10 artificial neural network system (ANNs) with three layers has bee n developed in this paper. The simultaneous concentration analysis of ten a ir toxic volatile organic compounds (VOCs) was resolved. The net was traine d with a back- propagation algorithm. The input signals were the absorbance values at the selected peak wavenumbers and a series of equispaced wavenum bers. The output is the predicted concentration for each component in the p rediction samples. Several parameters were optimized to ensure convergence and to speed up learning. The comparison of prediction results at the chara cteristic peak wavenumbers and the series of equispaced wavenumbers showed that results obtained at the peak absorption wavenumbers were superior to t hat at the equispaced wavenumbers. The results also showed that ANN can suc cessfully resolve the concentration analysis problem when the FTIR spectra of several constituents in the mixture interfere with each other.