Application of ANN to the Sorption Equilibrium Modelling of Heavy Metal Ions on Clinoptilolite

The latest achievements in the field of mathematical modelling include the application of artificial neural networks (ANN). A growing interest in the ANN is confirmed by the number of publications devoted to the applicability of ANN in chemical, process and environmental engineering. A recent dynamic development of ANN provided an efficient and universal tool that is used to solve many tasks, including modelling, approximation and identification of objects. The initial step of applying the network to a given process consists in the determination of weights of the proposed neural network structure. This is performed on the basis of training data. A network that is properly trained allows correct information to be obtained on the basis of other data which have not been used in the network training. In most cases the network training is performed on the basis of a known mathematical model. However, the training of a network can be also performed using experimental data. In this paper, the sorption isotherms were predicted by means of a multilayer perceptron (MLP). Calculations were made using a training program written in Matlab, which took advantage of the Lavenberg-Marquardt procedure. In the last decade a growing interest is observed in inexpensive and very cheap adsorbents to remove heavy metal ions. Clinoptilolite is the mineral sorbent extracted in Poland used to remove heavy metal ions from diluted solutions. Equilibrium experiments were carried out to estimate sorptivity of a clinoptilolite and its selectivity towards Cu(II), Zn(II) and Ni(II) ions for multicomponent solution. Calculations with the use of MLP enabled description of sorption isotherms for one, two and three ions which were present at the same time in the solution. The network also enabled an analysis of sorption of the single ion, taking into account the effect of its concentration.