The Speciation of Iron by a Wavelet Packet Transform Based Generalized Regression Neural Network

This paper presented a novel method named wavelet packet transform based generalized regression neural network (WPTGRNN) for studying the speciation of iron. The method combines wavelet packet thresholding denoising with generalized regression neural network. Wavelet packet representations of signals provided a local time-frequency description and separation ability between information and noise. The quality of the noise removal can be further improved by using best-basis algorithm and thresholding operation. Generalized regression neural network (GRNN) was applied for overcoming the convergence problem met in back propagation training and facilitating nonlinear calculation. In this case, the relative standard error of prediction (RSEP) for total compounds with WPTGRNN, WTGRNN, GRNN and PLS were 1.146, 1.865, 1.974 and 3.703 % respectively. Experimental results showed WPTGRNN method to be successful and better than others.