冷轧带钢板形控制矩阵机理智能模型研究

发布时间：2018-03-22 21:23

本文选题：板形　切入点：闭环控制　出处：《燕山大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着国民经济的发展和现代生活水平的提高,板带材的需求量在不断增加,同时对板带材产品质量的要求也日益提高。板形是带钢的重要质量指标,也是轧制领域研究的热点。近年来,人工智能方法以其在建模和控制方面的优势,在工业过程研究中得到了广泛的应用。首先,对一般的板形闭环控制的方法进行了分类和对比研究,阐述一般控制方法的原理和提出存在的问题和不足。其次,建立机理板形预报模型。为了分析各种板形调控手段和带钢来料情况对板形影响矩阵的影响规律,基于辊系弹性变形和金属模型相互耦合原理建立了板形机理预报模型,以1050六辊轧机为例,在此模型基础上计算机理方法影响矩阵,为板形在线控制模型的建立实现提供了理论依据。然后,建立智能板形预报模型。针对传统板形预报模型中采用的神经网络参数设置复杂,训练时间长,易陷入局部最小值缺点,引用极限学习机(ELM,extreme learning machine)人工神经网络用于板形预测,考虑到随机输入权值与偏置值对神经网络预测精度的影响,采用差分进化算法(DE,Differential Evolution)优化ELM神经网络,建立DE-ELM板形智能预测模型,在此模型基础上计算智能方法影响矩阵。最后,建立机理和智能加权综合影响矩阵控制模型。既发挥了机理模型的对轧制规律的把控,又利用智能模型对机理模型智能动态的修正,于是建立了机理-智能加权影响矩阵控制模型,并且在1050六辊轧机上进行了仿真,结果充分验证了本文提出的板形控制的动态影响矩阵法的有效性。
[Abstract]:With the development of national economy and the improvement of modern living standard, the demand for strip material is increasing, and the demand for product quality is also increasing. In recent years, artificial intelligence (AI) has been widely used in industrial process research because of its advantages in modeling and control. This paper classifies and contrasts the general method of shape closed loop control, expounds the principle of the general control method and puts forward the existing problems and shortcomings. Secondly, In order to analyze the influence of various shape control methods and strip feed on shape matrix, the prediction model of shape mechanism is established based on the coupling principle of roll elastic deformation and metal model. Taking the 1050 six-high rolling mill as an example, the influence matrix of computer method is based on the model, which provides the theoretical basis for the establishment and realization of the on-line control model of flatness. The intelligent shape prediction model is established. The neural network used in the traditional shape prediction model is complex in setting, long training time and easy to fall into the defect of local minimum value. The artificial neural network of extreme learning machine (ELM) extreme learning machine is used to predict the shape of shape. Considering the influence of random input weights and bias values on the prediction accuracy of neural networks, the differential evolution algorithm is used to optimize ELM neural networks, and an intelligent prediction model of DE-ELM flatness is established. Based on this model, the influence matrix of intelligent methods is calculated. The mechanism and intelligent weighted comprehensive influence matrix control model is established, which not only controls the rolling law of the mechanism model, but also uses the intelligent model to modify the intelligent dynamic of the mechanism model. A mechanism-intelligent weighted influence matrix control model is established and simulated on 1050 six-high rolling mill. The results fully verify the effectiveness of the dynamic influence matrix method proposed in this paper for shape control.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TG334.9

【参考文献】