[Studies on simultaneous quantitative determination of overlapping spectra using chemometric techniques].

Based on comparative study of eight chemometric denoising methods, a wavelet packet transform Elman recurrent neural network (WPERNN) method was developed to study simultaneous quantitative determination of overlapping spectra. The quality of noise removal and ability of regression were improved by combining wavelet packet transform with Elman recurrent neural network. Through optimization, the wavelet function, the wavelet packet decomposition levels as well as the structure and parameters of Elman recurrent neural network were selected. Two programs, PWPERNN and PERNN, were designed to perform WPERNN and ERNN calculation. Seven kinds of chemometric methods were applied in the present study for comparison. Experimental results showed that the WPERNN method was successful and better than the other 6 methods.