Research On Burn-through Point Prediction Model Of Sintering Process

The sintering process is an important process of the blast furnace. The sinter quality directly affects the blast furnace and steel production. Burn-through Point (BTP for short) is the position after the sintering, which is one of the reference for judging if the sintering process is normal or not. It is the key parameter closely related to the yield, quality and cost of sinter and the main basis for the operation of the sintering machine. Due to the characteristics of BTP lag in sintering process, the BTP prediction model needs to be established to predict the BTP values.The article introduces the process of sintering production in detail and analyzes the focused ignition sintering process which determines the sintering quality. As there is no device for the direct test of BTP, indirect measurement of BTP through other parameters is required. After a comprehensive analysis of several approaches for the calculation of BTP, the author adopted bellows exhaust temperature judgment method to measure the sintering end point indirectly and modified the final results. In view of the characteristics of sintering process, the relationship between sintering end point is established through the choice of four variables:ignition temperature, charge pressure, moisture content, the machine speed.In view of the characteristics of sintering process of multi variable, strong coupling, time-varying, nonlinear, the author modeled the BTP and established the BTP prediction NARMAX model.The variables of NARMAX order has an important influence on the accuracy of model, In order to obtain the established NARMAX model order, the theory of phase space reconstruction is introduced into two steps. First calculate the sampling interval of the variables and then seek for the model of variable order and then elaborated the process for determining the variable parameters of the mutual information method and pseudo proximal point algorithm, and determined the variable sampling interval and order the simulation analysis.With the experimental data, we need to identify the established BTP NARMAX model. As a machine learning, the support vector machine is able to approach nonlinear function with arbitrary precision and effectively overcomes the problems in the convergence of neural network which is unstable solution, poor generalization. Optimizing the kernel function of support vector machine and the penalty factors with the particle swarm optimization algorithm. Through the contrast experiment, the author confirms the rationality of the sampling interval based on mutual information and the determination of variable order based on pseudo proximal point method.