Hot Metal Pre-Desulfurization Prediction Model Based On Neural Network

Sulfur is a harmful element in most of steel. Hot metal pretreatment process is mainly used by Iron and Steel Plant domestic and overseas to reduce sulfur content in molten iron and increase steel quality. But the hot metal pretreatment process is still on manual control stage both home and abroad. For this reason, quite a few researchers proceeded a lot of work on desulfurization prediction model. However, these models were not suit for PZH Steel because they were created based on the special conditions of each plant.Contraposed the raw material and equipments of hot metal pretreatment in Vanadium Recovery and Steelmaking Plant of PSV, This paper built desulfurizer dosage prediction model using neural network modeling method, which was selected by analysis and comparison on desulfurization prediction model researched both domestic and overseas. Test was take place by using the historical date of PZH Steel. The results shown: the relative error of desulfurizer dosage between model prediction and actual value was less than 7 percents, the predicted desulfurizer dosage by model was comparative accurate.In the actual production process, conditions were not constant all the time. Considering the possibility of process mutations, it was not enough to just build desulfurizer dosage prediction model. This paper built feedback compensation model by using a sort of simple mathematic algorithm to quickly adjust desulfurizer dosage to face with the production in a short time. The on line training model was established for updating model parameters by using new production data. Meanwhile, based on the conditions of desulfurization production in PZH Steel, the blowing control parameters computation model was set up to provide reasonable control range of these parameters.To inspect the application effect, field trials were carried in production sites in PZH Steel. The results shown: end-[S] charging production according to model predicted value mainly met requirement of aim-[S]. The model could provide key parameters for desulfurization charging control system. Compare with former operation, the desulfuizer dosage used model was reduced 50～150 kilograms per furnace, and the injection time was decreased 2～3 minutes. It not only created the condition for desulfurization automation, but also had benefit to improve production efficiency and reduce the cost.