Prediction of pre-oxidation efficiency of refractory gold concentrate by ozone in ferric sulfate solution using artificial neural networks

Abstract

An artificial neural network model was developed to predict the oxidation of refractory gold concentrate (RGC) by ozone and ferric ions. The concentration of ozone and ferric ions, pulp density, oxygen amount, leaching time and temperature were employed as inputs to the network; the output of the network was the percentage of the ferric extraction iron from RGC. The multilayered feed-forward networks were trained by 33 sets of input-output patterns using a back propagation algorithm; a three-layer network with 8 neurons in the hidden layer gave optimal results. The model gave good predictions of high correlation coefficient ( R 2=0.966). The predictions by ANN are more accurate when compared with conventional multivariate regression analysis (MVRA). In addition, calculation with ANN model indicates that temperature is the predominant parameter and ozone concentration is the lesser influential parameter in the pre-oxidation process of refractory gold ore. The ANN neural network model accurately estimates the ferric extraction during pretreatment process of RGC in gold smelter plants and can be used to optimize the process parameters.