Research On Alloying Model For Ladle Furnace

Ladle furnace has become one of the key equipments in secondary refining process. It plays an important role in the control of components of molten steel, especially the alloy components. Now, the research on alloying component controlling has become one of the main subjects in ladle furnace research area. In the refining process of Ladle furnace, according to the requirements of different types of steel, we need add alloy to adjust the composition of liquid steel to make liquid steel achieve the target composition. Because each alloy is composed of several different elements and different alloys may have the same elemental composition, there are several ingredients programs. Therefore, in order to realize requires of component goal and the minimization of alloying cost, this paper uses the simplex method to establish the best ingredients model to make steel liquid composition achieve the standards. At the same time, it optimizes alloy charging scheme reasonably and achieves the purpose of controlling steel liquid ingredients and reducing alloying cost.However, the precision of alloying model relies heavily on computation of recovery rate of alloying elements. Therefore, in order to obtain accurate alloying model, we must calculate the recovery rate of alloying elements accurately. In the refining process of Ladle furnace, there are many factors that affect recovery rate of alloying elements, such as the steel temperature, oxygen level and so on, their relationship is very complex. As such a system with multi-variable, time-varying, nonlinear, coupled, large inertia, large time delay characteristics, the conventional modeling methods can not achieve the desired results.Therefore, this article uses the BP neural network to establish the model for recovery rate of alloying elements to overcome the difficulties of modling.On the basis of prediction model established by the BP neural network, introducing the SPSO to optimize this model and improve the convergence rate of BP neural network. PSO algorithm is simple and easy to implement. Its convergence speed is very fast in early to improve the convergence rate and the convergence accuracy of BP neural network. But the PSO algorithm will be affected by random oscillations in late to make it take a long time to search the global optimal value, its convergence speed is low, it limit to local minimum easily to lead to low precision and divergence. Therefore, on the basis of PSO, we introduce SA algorithm to overcome shortcomings that the algorithm limits to local minimum easily and we use MATLAB software for simulation experiment of BP network, PSO optimize BP network and SA-PSO optimize BP network, and the simulation results are compared.