Research Of Temperature Prediction Model In The Process Of RH Refining

This paper takes RH refining process of a steel mill as the research object. First, it established mechanism model and analyzed the factors affecting the temperature of molten steel in the model with the data obtained by the sensor system. It summed up the main factors influencing the temperature of molten steel are the weight of the molten steel, the initial temperature of the molten steel, refining time, the oxygen content of the molten steel and the amount of added aluminum. Compared to the method of calculation of human experience, this method has a more adequate scientific basis and more accurate results. Then, according to the types of steel and RH refining process, it divided the models into temperature prediction model of molten steel in VCD time and temperature prediction model of molten steel in the final time. The inputs of the temperature prediction model of molten steel in VCD time are the weight of the molten steel, the initial temperature of the molten steel, the initial oxygen content of the molten steel and the VCD time; the output is the temperature of molten steel in VCD time. The inputs of temperature prediction model of molten steel in the final time are the weight of the molten steel, time of VCD, the temperature of molten steel in VCD time, the oxygen content of the molten steel in VCD time, the final time, the amount of added aluminum; the output is the temperature of the molten steel in the final time. Then, it uses the method of stepwise regression to find the input that has a significant impact on the output in temperature prediction model. This method does not require a deep understanding of mechanism of metallurgy and can shorten the modeling cycle, and the results of its analysis are coincide with the factors summarized by mechanism model. Finally, it established a neural network model, took the actual metallurgical data as the sample, used particle swarm algorithm to train the neural network. This method has a fast training speed and a strong ability of convergence.It simulates the model using Matlab and the simulation results show that nine hidden layer neurons would be appropriate in temperature prediction model of molten steel in VCD time. The accuracy with an error of ±6? reached 92%; sixteen hidden layer neurons would be appropriate in temperature prediction model of molten steel in the final time, and the model's accuracy with an error of ±5? reached 95%. Therefore, compared to the traditional temperature prediction mechanistic model and the neural network model, the temperature prediction model in this paper has a higher prediction accuracy and a stronger ability of generalization.