Improvement of Prediction Ability of Multicomponent Regression Model

Abstract

A novel method named wavelet packet transform based Elman recurrent neural network (WPTERNN) was proposed for simultaneous determination of Co (II), Zn (II) and Cu (II) by combining wavelet packet denoising with Elman recurrent neural network. Wavelet packet representations of signals provided a local time–frequency description, thus in the wavelet packet domain, the quality of the noise removal can be improved. Elman recurrent network was applied for non-linear multivariate calibration. In this case, by trials, wavelet function, decomposition level and numbers of hidden nodes for WPTERNN method were selected as Daubechies 2, 3 and 9 respectively. A program PWPTERNN was designed to perform simultaneous determination of Co (II), Zn (II) and Cu (II). The relative standard errors of prediction (RSEP) for all components with WPTERNN, Elman recurrent neural network (ERNN) and partial least squares (PLS), principal component regression (PCR) and Fourier transform based PCR (FTPCR) were 6.7, 14.7, 9.2, 25.6 and 25.2 % respectively. Experimental results demonstrated that the WPTERRN method had the best performance among the five methods.