End‐Point Dynamic Control Model for 260 Tons BOF Steelmaking Based on PWWTSVR and LWOA

Abstract

The end-point carbon content and temperature are two important factors that affect steel quality in the basic oxygen furnace (BOF) production process. [1] The BOF end-point control [2] can affect the quality of subsequent products and production costs of expenditure. Therefore, the research on BOF end-point control is particularly important. At present, some mathematical models can be established to realize the control of the BOF end-point, which are the static control model and dynamic control model. The static control model is based on the initial smelting conditions, using mechanisms, [3] statistics, [4] and other algorithms to establish a model for related calculations. Due to the changes in the blowing process that cannot be monitored and adjusted in time, and the accuracy cannot be improved. The dynamic control model is based on the static control model, which is mainly to dynamically monitor the blowing process in the later stage of smelting to adjust the parameters. Compared with the static control model, the dynamic control model has higher accuracy. Therefore, it is very meaningful to deeply investigate the dynamic control model. The premise of establishing a dynamic control model is based on an accurate prediction model, and the results of the prediction model can affect the proposed control model. Therefore, it is necessary to establish an accurate BOF end-point prediction model. With the application development of artificial intelligence technology and computer network technology, the neural network has been applied to BOF end-point prediction by many scholars. [5][6][7][8][9][10] Cox et al. [5] established a converter end-point prediction model based on the artificial neural network, which was used to predict the oxygen blowing volume and coolant addition at the end of the blowing period. Jim'enez et al. [6] used an artificial neural network to establish a prediction model to predict the temperature of molten iron. Fileti et al. [7] established an inverse neural model to adjust the oxygen and coolant required for the end-blow stage. Wang et al. [8] proposed a prediction model based on the combination of a genetic algorithm and back propagation (BP) neural network and verified that the combined model has a high end-point hit rate (HR). Wang et al. [9] combined a weighted clustering algorithm with a polynomial neural network to establish a converter end-point prediction model, which could effectively predict the end-point phosphorus content of molten steel. He et al. [10] used principal component analysis to reduce the dimension of the input variables of the prediction model, and then used BP neural network to establish a prediction model for the end-point phosphorus content of the converter. The previous research has achieved good results. However, the neural network may fall into the local optimal problem when establishing the prediction model, which could not improve the accuracy of the model. Support vector regression (SVR) is an intelligent algorithm for regression problems proposed by Cortes et ...