基于PSO-BP神经网络的双机架炉卷轧机轧制力预测

发布时间：2018-03-18 02:38

本文选题：双机架炉卷轧机　切入点：轧制力　出处：《昆明理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：轧制过程数学模型是现代轧机计算机控制系统的重中之重,轧制力数学模型对钢铁产品质量和厚度精度等有着不可替代的作用。当今各行业对钢铁产品的质量要求愈发严格,怎样提高轧制产品的质量是一个亟待解决的重要问题。双机架炉卷轧机轧制力的精度影响着轧制产品的质量。而轧制区金属的塑性变形是一个非常复杂的过程,轧制过程中的很多参数与双机架炉卷轧机轧制力存在非线性的强耦合关系。根据传统数学模型公式和研究经验远不能满足当前轧制力精度的高要求,也不能对双机架轧机中轧制力的变化过程进行准确的描述。所以本文利用轧机的轧制压力数学模型将神经网络与智能算法来相结合来进行轧制力的高精度预测,这也是当前研究认为最有效的预测途径。本文以轧制压力数学模型的变形理论为基础,以某大型轧钢厂1725mm双机架轧机轧制压力模型为研究对象,结合西姆斯(SIMS)公式着重分析了轧制变形区的基本工艺参数、轧制前后的基本规律等,确定了对轧制力计算精度影响较大的参数变量,这些参数变量主要有轧件厚度、轧件温度、轧辊半径、轧制速度等;然后在轧制力模型的基础上,以影响轧制力精度的主要变量作为BP神经网络的输入量,以轧制力作为输出量建立了 BP网络拓扑结构为10-12-1的轧制力预测模型。根据该大型钢厂1725mm双机架轧机轧轧制钢种为Q235的实测数据对所建立的BP网络模型进行训练和测试。利用训练好的神经网络对轧机轧制力进行预测,然后分别使用粒子群算法、改进的粒子群算法来对BP神经网络进行优化,再次对两种预测模型进行轧制力预测,仿真结果表明经改进的PSO融合算法的轧制力预测性能达最佳的逼近效果,PSO优化BP网络的算法预测性能次之。因此,本文最终建立了基于PSO-BP神经网络的双机架炉卷热轧机轧制力预测模型,该模型有效提高了双机架炉卷轧机轧制力的预测精度。
[Abstract]:The mathematical model of rolling process is the most important part of the computer control system of modern rolling mill. The mathematical model of rolling force plays an irreplaceable role in the quality and thickness accuracy of steel products. How to improve the quality of rolling products is an important problem to be solved urgently. The precision of rolling force of double-stand Steckel mill affects the quality of rolled products. The plastic deformation of metals in rolling area is a very complicated process. There is a nonlinear strong coupling relationship between many parameters in rolling process and rolling force of double stand Steckel mill. According to the traditional mathematical model formula and research experience, it is far from meeting the high requirement of rolling force precision at present. The rolling force change process in the double stand rolling mill can not be accurately described, so this paper combines the neural network with the intelligent algorithm to predict the rolling force with high precision by using the rolling force mathematical model of the rolling mill. This is also the most effective method for prediction. Based on the deformation theory of the mathematical model of rolling force, this paper takes the rolling pressure model of 1725mm double stand mill in a large rolling mill as the object of study. Combined with Simsberg SIMS formula, the basic technological parameters of rolling deformation zone and the basic rules before and after rolling are emphatically analyzed, and the parameter variables which have a great influence on the calculation accuracy of rolling force are determined. These parameters mainly include the thickness of rolling piece, the temperature of rolled piece, etc. Based on the rolling force model, the main variables affecting the precision of rolling force are used as the input amount of BP neural network. The rolling force prediction model with BP network topology structure of 10-12-1 was established with rolling force as output. The BP neural network model was trained according to the measured data of 1725mm double-stand mill rolling and rolling steel grade Q235. Using trained neural network to predict rolling force of rolling mill, Then the BP neural network is optimized by using particle swarm optimization algorithm and improved particle swarm optimization algorithm, and then the rolling force of the two prediction models is predicted. The simulation results show that the improved PSO fusion algorithm can achieve the best predictive effect of rolling force. In this paper, a rolling force prediction model based on PSO-BP neural network is established, which can effectively improve the precision of rolling force prediction of double-stand Steckel rolling mill.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG333;TP183

【参考文献】