An Neural Network Model for the Fe/SiO2 Ratio in Copper Flash Smelting Slag Using Improved Back Propagation Algorithm

The Fe/SiO2 ratio in slag is one of the important control parameters for copper flash smelting process, but it is difficult to describe the complex relationship between the technological parameters and the Fe/SiO2 ratio in slag using accurate mathematic formulae, because the copper flash smelting process is a complicated nonlinear system. An neural network model for the Fe/SiO2 ratio in copper flash smelting slag was developed, whose net structure-is 8-15-12-1, and input nodes include the oxygen volume per ton concentrate, the oxygen grade, the flux rate, the quantity of Cu, S. Fe, SiO2 and MgO in concentrate. In order to avoid local minimum terminations when the model is trained by back propagation (BP) algorithm, a new algorithm called GA-BP is presented by using genetic algorithm (GA) to determine the initial weights and threshold values. The results show that the model can avoid local minimum terminations and accelerate convergence, and has high prediction precision and good generalization performance. The model can be used to optimize the copper flash smelting process control.