Operation Analytics And Optimization Of Typical Process In The Iron&Steel Production

With the requirements of high product quality and environmental protection, there is urgent need for the study of the estimation, prediction, diagnosis and control problems existing in the typical iron&steel production processes, designing algorithms conforming to the problem characterisitics and improving the control accuracy and product quality. Five typical processes in iron&steel production are studied in this dissertation, and data analytics and operation optimization menthod are utilized to solve the estimation, prediction, diagnosis and control problems. The main contents of the dissertation are summarized as follows:(1) The cross temperature of blast furnace is related with dozens of input variables such as the flow and pressure of hot air and blast furnace permeability index, and that the mechanism of action of the input variables are not clear. Therefore, it is difficult to obtain accurate mechanism model. To obtain accuracy model of the cross temperature of blast furnace, LS-SVM based on PSO is designed to build the model based on production data quickly. Firstly, the correlation analysis is done to select inputs related to cross temperature; Secondly, an improved PSO is proposed to obtain optimized parameters for LS-SVM to improve the prediction accuracy; Finally, the prediction model of the cross temperature in blast furnace based on LS-SVM is achieved. The experiments illustrate that the accuracy of the proposed data-driven cross temperature model occupies the improvement of 3% compared with grid search method, and can meet the requirements of production.(2) The traditional detection method of the steel coil temperature in the bell type annealing furnace is easy to damage the coil by using buried thermocouple. To solve the detection problem of steel coil temperature in the bell-type annealing furnace, an improved PSO-based adaptive LS-SVM is proposed to obtain the accuracy estimation model. In this algorithm, an adaptive kernel function that based on theoretical analysis aimed at surpassing the limitations of existing kernels is developed. An improved PSO with a local search strategy optimizes the adaptive kernel parameters dynamically for different problems. The experimental results illustrate that the PSO-based adaptive LS-SVM can achieve higher accuracy and good generalization ability. And the proposed method can detect the coil temperature with high accuracy without coil damage which meets the acquirement of actual production.(3) To meet the characteristics of high-intensity signal noise, high-accuracy classification requirement and high-speed production process in the steel industry, the tailed and improved support vector machine (TISVM) is proposed in combination with the improved PSO and logistic regression (LR). During the implementation of the TISVM, the adaptive feature weighting strategy reduces the influence of noise on the defect images. At the same time, the IPSO sets the TISVM parameters dynamically to increase the classification accuracy. To speed up the TISVM, LR filters out unpromising particles during iterations of the IPSO. The experimental results illustrate that IPSO-based parameter optimization and feature weighting are effective for increasing the classification accuracy, and LR improves the computational efficiency of TISVM significantly.(4) Since the heating furnace of continuous annealing processing has the characteristics of high-speed production process and production diversification, the traditional methods can hardly control the strip temperature accurately. Therefore, a PSO based operation optimization method is designed to solve this problem. Firstly, the mechanism model describes dynamic relationship between the strip temperature and the variables that include the fuel flow rate and the strip width, the thickness and the speed of the strip. Based on the model of strip temperature the optimization model is constructed. It aims at both the minimum of the error between strip temperature and the ideal value and the minimum of the fuel consumption. Then the model is solved by improved PSO algorithm. The results of the experiment illustrate that the proposed method can adapt to the high-speed production process, control the strip temperature accurately, and reduce energy consumption.(5) Since the production process has the characteristics of high temperature, high noise, large time-delay and complex mechanism, it is difficult to utilize tradiction model predictive control to this complex process. To control the steel slab temperature of reheating furnace process with high quality, a two-stage PSO-based nonlinear model predictive control (NMPC) method is proposed. In this method, PSO is utilized to optimize the parameters of SVM for different problems to construct the nonlinear predictive model based on the real production data. Then PSO solves the rolling optimization problem in NMPC to obtain the proper control variables. The experiment results illustrate that the PSO-based SVM can obtain accurate predictive model. Moreover, the proposed nonlinear model predictive control method can obtain outstanding control accuracy in steel slab temperature control with average error under 3% which meet the actual requirement.