基于改进神经网络的热轧厚度控制方法研究

发布时间：2018-09-02 07:22

【摘要】：钢铁的产量和质量是衡量一个国家工业化发达程度的重要指标。钢铁工业是工业化发展的基础产业。随着现代生产水平的提高许多行业，例如汽车制造、航空航天以及房地产等对带钢板质量要求也不断提高。由于带钢板的厚度精度是衡量带钢板质量的最重要指标。所以控制带钢厚度，提高带钢厚度精度成为国内外冶金行业普遍关注的重要课题。厚度自动控制系统又称为AGC（Automatic Gauge Control）系统，它是一个多变量、强耦合的非线性控制过程，目前大多以现代控制技术的应用和人工智能算法的引用来实现控制性能的最高水平，提高控制系统的灵活度和方便性。目前在带钢实际生产中热轧厚度控制系统是带钢厚度控制领域应用最广泛的控制系统。本文先是介绍了热轧带钢生产工艺和流程，以及厚度控制的理论基础。研究了厚度控制的基本数学模型和影响厚度波动的原因。然后介绍了几种AGC的控制原理。通过分析液压AGC的主要组成部分控制器、伺服放大器、伺服阀、液压缸和传感器，并为各组成部分建立数学模型，建立起热轧AGC的数学模型，由于研究重点在于控制器的自适应调整，，因此对复杂的数学模型进行降阶处理，将复杂的五阶模型简化为二阶数学模型。本文研究的重点是神经网络的改进与其在热轧厚度控制中的应用。由于神经网络具有的高度非线性信息处理能力和自学习能力，使它可以逼近任意的高度非线性系统，但是神经网络在某些情况下存在一些缺陷，如收敛速度缓慢，易陷入局部极小值等。因此如何利用神经网络的优点，克服它的缺陷，并对其进行改进和创新，用来解决热轧厚度控制的问题将是我们研究的重点。本文针对这些缺陷将神经网络进行优化处理，引入了改进神经网络的具体方式：添加动量项。本文还研究探索了对神经网络结构的选取、非线性误差函数的选择以及动量因子选取等方面。通过仿真实验确定其确定了一种跟踪性能良好的BP神经网络结构，并通过仿真实验验证了这种结构的神经网络的有效性。最后本文设计了将BP神经网络改进后的PID控制器，并将此控制器应用在热轧厚度控制系统中，通过仿真实验结果证明它具有良好的跟踪性能和信息处理能力。
[Abstract]:The output and quality of iron and steel is an important index to measure the degree of industrialization of a country. Iron and steel industry is the basic industry of industrialization development. With the improvement of modern production level, many industries, such as automobile manufacturing, aerospace and real estate, also improve the quality requirements of steel plate. Because the thickness accuracy of strip plate is the most important index to measure the quality of strip plate. Therefore, controlling strip thickness and improving the precision of strip thickness have become an important topic of general concern in metallurgical industry. Thickness automatic control system, also known as AGC (Automatic Gauge Control) system, is a multivariable, strongly coupled nonlinear control process. At present, most of the applications of modern control technology and the introduction of artificial intelligence algorithm to achieve the highest level of control performance. Improve the flexibility and convenience of the control system. At present, the thickness control system of hot rolling is the most widely used control system in the field of strip thickness control. This paper first introduces the production process and process of hot strip, and the theoretical basis of thickness control. The basic mathematical model of thickness control and the reason of influencing thickness fluctuation are studied. Then several control principles of AGC are introduced. By analyzing the main components of hydraulic AGC controller, servo amplifier, servo valve, hydraulic cylinder and sensor, and establishing mathematical model for each component, the mathematical model of hot rolling AGC is established. Because the emphasis of the research is on the adaptive adjustment of the controller, the complex mathematical model is reduced to a second-order mathematical model. The emphasis of this paper is on the improvement of neural network and its application in the thickness control of hot rolling. Because of the highly nonlinear information processing ability and self-learning ability of neural network, it can approach any highly nonlinear system. However, in some cases, the neural network has some defects, such as slow convergence rate. Easy to fall into local minima, etc. Therefore, how to make use of the advantages of neural network, overcome its defects, and improve and innovate it to solve the problem of hot rolling thickness control will be the focus of our research. In this paper, the neural network is optimized for these defects, and the specific way to improve the neural network is introduced: adding momentum term. The selection of neural network structure, the selection of nonlinear error function and the selection of momentum factor are also studied in this paper. The BP neural network structure with good tracking performance is determined by simulation experiments, and the effectiveness of the neural network is verified by simulation experiments. Finally, an improved PID controller based on BP neural network is designed and applied to the thickness control system of hot rolling mill. The simulation results show that the controller has good tracking performance and information processing ability.
【学位授予单位】：济南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TG334.9

【参考文献】