Study on Smelting Reduction of Coal-Containing Pellets of V-Ti Bearing Beach Placers by Combined Rotary Hearth Furnace and Direct Current Arc Furnace

"The smelting reduction of V-Ti bearing beach placer by combined rotary hearth furnace and direct current arc furnace was studied in laboratory. It takes the aid of back propagation (BP) neural network theory to build the nonlinear mapping relations between the crucial process variables such as content of carbon, temperature and time and the degree of reduction. Then by the integrating BP neural network and genetic algorithm (GA), the optimized process parameters for the high degree of reduction were searched. The comparisons between experiment results and neural network simulation results show that GA-based on BP method can predict the degree of reduction with higher prediction accuracy. Calculations show that the integrated energy consumption of new technology is 580kgce/tHM, less than the current existing blast furnace.IntroductionThese V-Ti bearing beach placers (VTBBP) are secondary Fe-rich minerals formed by rivers, ocean waves, ocean tides and ocean currents on beach zones, their useful minerals are V-Ti magnetite ores. On Asia Pacific zone, such as Japan, China, Philippines, Indonesia, Australia and New Zealand, the VTBBP have large reserves and wide distribution. But so far, the VTBBP are not still comprehensively utilized. Therefore, in this paper, a combined rotary hearth furnace technology and arc furnace smelting method is studied for comprehensive utilization of the VTBBP. The main objective of this paper is to find optimum operation conditions for the production ofV-bearing iron and titania-rich slag from the VTBBP by means of the laboratory testing.Recently, neural network (NN) expected to be able to provide an effective tool because of its advantages, which describe nonlinear mapping relations. With the characteristics of strong selflearning, self-organization, robust error toleration and accurate nonlinear relation approximation, artificial neural network can be applied to nonlinear process modeling based on sufficient training. Back-propagation (BP) training algorithm is probably the frequently used one in practical application. Genetic algorithm has parallel search strategy and global optimization characteristics, which makes the trained neural network being higher classification accuracy and faster convergence speed. So it is necessary to combine neural network and genetic algorithm. The nonlinear relationship between input and output presented by NN and the global optimal function of GA are abroad applied in the engineering and scientific research [1, 2]. So these methods supply an efficient path to solve the above problems. At present, it is common to study the optimization of process parameters by orthogonal experimental design, the method combining neural network and genetic algorithm is a novel and better method than orthogonal experimental design. Tills study organized the above advantages to optimize the process parameters for comprehensive utilization of the VTBBP."