Abstract:

A BP artificial neural network (ANN) model was developed via MATLAB to estimate Cu/Zn adsorption in the surficial sediments (SSs)
and natural surface coating samples (NSCSs), by which the mean square error of training set, the deviation of training set, the mean square error of verification set, and the deviation of test set are 0.002 2(0.001 5), 1.542 9×10^-6(2.648 4×10^-6), 0.087 1(0.069 2), and 0018 7(0.035 7), respectively. The predicted results of the BP ANN model established not only reveal the law of the adsorption of Cu/Zn by the active constituents of SSs (NSCSs) changed in grade, but also highlight the significant interaction among the activeconstituents in the SSs (NSCSs) on the adsorption of Cu/Zn preliminary. The adsorption capacity of Cu/Zn increases with the decrease in content of active constituents, and the maximum ratios between the adsorption variation value of the changed SSs (NSCSs) and their original adsorption value are 1 420.30%/54.30% (SSs) and 79.27%/703.31% (NSCSs) respectively; the interaction among the active constituents in SSs (NSCSs) promote Cu/Zn adsorption, and the maximum ratios between the adsorption variation value of the changed SSs (NSCSs) and their original adsorption value are 386.14%/30.08% (SSs) and 66.17%/47.92% (NSCSs) respectively.

Key words: surficial sediments, heavy metals, adsorption, BP artificial neural network, interaction