| EIA Home | EIA 2004 | Subscription |
International Society for Environmental Information Sciences
Environmental Informatics Archives
ISSN 1811-0231 / ISEIS Publication Series Number P002
Copyright © 2004 ISEIS. All rights reserved.
Guest
Paper EIA04-071, Volume 2 (2004), Pages 711-721 = complimentary
Radial Basis Function Neural Networks in the Analysis of Polycyclic Aromatic Hydrocarbons by Synchronous Fluorescence
G. Zheng, W. H. Huang* and X. H. Lu
School of Environmental Science and Engineering, Huazhong University of Science and Technology, 1037 Luoyu Road, 430074 Wuhan, People’s Republic of China. *Corresponding author: whuang48@uwo.ca.
Abstract
Radial basis function neural networks (RBFNs), back-propagation neural networks (BPNs) and partial least squares (PLS) have been compared in order to establish the best multivariate models for the analysis of mixtures of polycyclic aromatic hydrocarbons (PAHs) containing 3 of these compounds (pyrene, benzo[a]anthracene and chrysene). The synchronous fluorescence spectra (recorded at wavelength increments of 100,115 and 150 nm) of 29 standards (20 standards as calibration set and 9 standards as validation set) have been used for this purpose. RBFN model with 4 neurons in the input layer and 4 neurons in the radial basis layer, BPN model with 4 neurons in the input layer and 3 neurons in the hidden layer and PLS model with 3 components have been obtained. RBFN and BPN models have performed better with lower root mean square errors of 1.15 and 0.99 respectively in contrast to 2.23 for PLS model. RBFN model has predicted the concentrations of 3 PAHs with nearly the same accuracy while BPN and PLS models have performed better for the PAHs with higher fluorescence intensity. With predicting accuracy comparable to BPN model, RBFN model needs only 4 epochs of training in contrast to 104 epochs of training for BPN model. Furthermore, it is time-consuming and empirical to optimize the number of neurons in the hidden layer of BPN model while the number of neurons in the radial basis layer of RBFN model is optimized automatically during the training of the network.
Keywords: PAH, Synchronous fluorescence, RBFN, BPN, PLS