Research On The Prediction Neural Network Of Fluidized Roasting For Changeful Zinc Concentrate

Zinc is one of the important nonferrous metals. China is the world's leading zinc producer. Nowadays, the consumption and production of zinc are respectively increasing year by year. Zinc smelting industries are generally facing a conflict between the increasing demand for zinc and the shortage of zinc concentrate. Changeful and high miscellaneous zinc concentrate will become the main raw materials of zinc smelting. The use of Changeful Zinc Concentrate will have a direct impact on the production conditions of zinc hydrometallurgy process. Fluidized roasting is an important process during the Zinc Smelting. The operating conditions such as roasting temperature, roasting time, the ratio of wind to material and the exports' pressure of fumes will have an important impact on the quality of zinc calcine. It is related to normal operation of the whole zinc hydrometallurgy process.Firstly, the main existing problems, new progress and the developmental orientation of zinc hydrometallurgy process are discussed in this paper. Then there is a focus on the changes of fluidized roasting process for zinc and associated elements. The thermodynamics and kinetics of fluidized roasting process are studied, and the Mechanism of zinc ferrite formation is analyzed. It points out that the formation of zinc ferrite is the main factor which affects the quality of zinc calcine. At the same time a way to reduce zinc ferrite generation is proposed.Based on collecting samples and data of roasting production on the production site and the theoretical analysis, the influence factors for the rate of soluble zinc, soluble iron, soluble sulphur and soluble silicon dioxide are studied. When the Zinc Concentrate have complex sources, the prediction BP neural network between material constituent, technological conditions and above zinc calcine are established. The BP neural network can predict the quality of zinc calcine, and can play a guide role in adjusting the roasting process operating parameters. There is no determined theory or method in choosing the number of hidden layer nodes in neural network. In this paper an optimization algorithm based on the principle of golden section is adopted to design the number of hidden layer nodes in neural network. By adopting traingdm function (momentum gradient drop the back-propagation algorithm) to train BP network, by comparing the predictive indexes with industrial data, and by drawing a conclusion that the predictive effect is good, neural network can be predicted accurately to set up prediction network.By designing the BP neural network the problem of hard establishment for mathematical model between technological conditions and the quality of zinc calcine is solved. So it is a meaningful attempt at zinc fluidized roasting process.