基于GA-BP神经网络模式识别的连铸机漏钢预报模型研究

发布时间：2018-03-11 08:53

  本文选题：连铸　切入点：粘结漏钢　出处：《大连理工大学》2015年硕士论文　论文类型：学位论文

【摘要】：连铸是现代炼钢企业铸造钢坯最主要的方法,它具有简化生产工序、钢水利用率高、耗能低、钢坯质量好、改善劳动条件、生产过程机械化和自动化程度高等优点。连铸技术的应用彻底改变了铸造车间的生产工艺流程、人员分配和物流控制,为钢铁生产的自动化和信息化技术的应用,从而大幅改善环境和提高产品质量提供了有力的物质条件保障。漏钢是连铸过程中极具危害性的重大生产事故,为避免连铸漏钢事故的发生,国际上通常从两个方向上着手研究解决,一方面是改善连铸设备和工艺条件,从漏钢形成的机理上杜绝事故的发生；另一方面是及时检测识别出漏钢的特征,采取减速拉坯等有效措施避免漏钢。本文对粘结漏钢的形成机理,漏钢模式识别原理进行了分析,理解温度的动态传递是漏钢过程的物理特性的基本映射。把热电偶的动态数据转化为静态数据,组合空间网络结构以便于区域判别法的实现。在此基础上寻求一种有效的基于GA-BP神经网络模式识别的连铸机漏钢预报模型。本文对结晶器粘结漏钢时热电偶温度的变化进行了分析,将热电偶的温度随时间变化规律作为模式识别的特征,从而进行是否漏钢的判定。本模型既建立了单个电偶的单偶时序网络,又建立了多个电偶的组偶空间网络,在组偶空间模型的建立中,区别之前的“1”字型二电偶空间网络结构和“⊥”字型的四电偶空间网络结构,建立了“T”字型的四电偶空间网络结构,并对两种四电偶空间网络结构进行了比对,因而连铸机漏钢模式的识别将更准确、更及时。本文对BP神经网络和遗传算法进行了探讨,将遗传算法与BP神经网络结合,利用遗传算法对BP神经网络的初始权值进行全局最优搜索,其中使用了利用导数修正种群中个体数的参数,从而改善了遗传算法局部搜索能力不强的缺点,加速了遗传算法的收敛速度,搜索到条件最优解后,将最优解赋予BP网络进行精确求解。这样既避免了局部极小的问题,又达到了快速高效和全局寻优的目的,使训练结果得到极大的改善。BP神经网络以最速下降法为学习准则,以误差反向传播算法进行连接权值的调整,将大量训练样本数据输入到神经网络中,得到最优的权值解,进而获得相对较优的连铸机漏钢预报模型神经网络。利用Visual Studio 2010集成开发环境,用C++语言完成程序的开发和界面设计,界面部分采用MFC设计完成。在软件开发中使用了具有跨语言、跨平台的专业图形程序接口功能的OpenGL完成热电偶波形的显示。训练并离线测试了单偶时序网络和组偶空间网络预报模型,测试结果表明,基于GA-BP神经网络模式识别的连铸机漏钢预报模型具有很好的非线性映射精度和分类识别能力。
[Abstract]:Continuous casting is the most important method for casting billets in modern steelmaking enterprises. It has the advantages of simplifying production process, high utilization ratio of molten steel, low energy consumption, good quality of billets and improving working conditions. The application of continuous casting technology has completely changed the production process, the distribution of personnel and the control of logistics in the foundry workshop, for the application of automation and information technology in iron and steel production. Thus greatly improving the environment and improving the quality of products provide a strong material condition guarantee. Steel breakout is a very harmful production accident in the continuous casting process, in order to avoid the occurrence of continuous casting breakout accidents, The international studies and solutions are usually carried out in two directions. On the one hand, it is to improve the equipment and technological conditions of continuous casting, to prevent the occurrence of accidents from the mechanism of steel breakout formation; on the other hand, to detect and identify the characteristics of steel breakout in time. In this paper, the forming mechanism of bond breakout and the principle of pattern recognition of steel breakout are analyzed. Understand that the dynamic transfer of temperature is the basic mapping of the physical properties of the breakout process. The combined spatial network structure is convenient for the realization of zone discriminant method. On this basis, an effective model for predicting steel breakout in continuous casting machine based on GA-BP neural network pattern recognition is sought. The thermocouple temperature of mould bonding breakout is studied in this paper. Has been analyzed, The temperature variation of thermocouple with time is taken as the characteristic of pattern recognition, so as to determine whether the steel breakout is broken. This model not only establishes the single pair sequential network of single couple, but also establishes the even space network of multiple pairs. In the establishment of the model of even space, the network structure of "1" type of two-electric couple space and "Karabakh" type of four-pair space is distinguished, and the "T" type of four-pair space network structure is established. Two kinds of four-couple spatial network structure are compared, so the recognition of the breakout pattern of continuous casting machine will be more accurate and timely. This paper discusses BP neural network and genetic algorithm, and combines genetic algorithm with BP neural network. The genetic algorithm is used to search the initial weights of BP neural network, in which the parameters of individual number in the population are modified by using the derivative, which improves the weak local search ability of the genetic algorithm. The convergence speed of genetic algorithm is accelerated. After searching the conditional optimal solution, the optimal solution is assigned to the BP network to solve the problem accurately. This not only avoids the problem of local minima, but also achieves the purpose of fast and high efficiency and global optimization. The training results are greatly improved. BP neural network takes the steepest descent method as the learning criterion, adjusts the link weight by the error back-propagation algorithm, and inputs a large number of training sample data into the neural network to obtain the optimal weight solution. Finally, a relatively good neural network for predicting steel breakout of continuous casting machine is obtained. Using Visual Studio 2010 integrated development environment, C language is used to complete the program development and interface design. The interface is designed by MFC. In software development, cross-language is used. The display of thermocouple waveform is accomplished by OpenGL with the function of professional graphic program interface across platforms. The prediction models of single and even sequence networks and even space networks are trained and tested offline. The test results show that, The failure prediction model of continuous casting machine based on GA-BP neural network pattern recognition has good nonlinear mapping accuracy and classification recognition ability.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TF777

【参考文献】

中国期刊全文数据库 前10条

1 何涛焘;田陆;黎锋果;刘琼;刘晓玲;王石雄;;基于热电偶测温的漏钢预报系统[J];传感器世界;2010年11期

2 罗婷;周新志;;GA-BP网络在去除MODIS遥感影像条带噪声中的应用[J];测控技术;2010年06期

3 张松艳;;选择算子与遗传算法的计算效率分析[J];宁波大学学报(理工版);2009年03期

4 李同彬;姚若华;陈波;;几种常用漏钢预报系统模型的比较(下)[J];冶金自动化;2009年04期

5 金焱;毕学工;;粘结漏钢裂纹扩展角度研究[J];南方金属;2008年04期

6 史聚建;;板坯连铸漏钢预报系统的研究[J];重庆科技学院学报(自然科学版);2008年01期

7 范建东,王唯一,荣亦诚,龚幼民;RBF神经网络应用于连铸漏钢预报[J];上海大学学报(自然科学版);2001年05期

8 王唯一,荣亦诚,龚幼民,范建东,王成虎,耿海,傅正权;基于神经模糊技术的漏钢诊断预报模型的研究[J];自动化仪表;2001年10期

9 职建军,文昊,裴云毅;宝钢板坯连铸漏钢预报系统(BBPS)的开发与应用[J];炼钢;2001年03期

10 胡婕;结晶器漏钢预报系统[J];山西机械;2000年S2期

中国硕士学位论文全文数据库 前6条

1 张水仙;板坯连铸过程漏钢预报神经网络算法研究[D];燕山大学;2012年

2 朱敏;基于自适应遗传BP神经网络的文本分类方法研究[D];南昌大学;2010年

3 杨吉林;基于GA-BP神经网络漏钢预报模型研究[D];东北大学;2010年

4 刘永贞;基于人工智能与热成像的结晶器漏钢预报系统[D];大连理工大学;2009年

5 吴展遥;连铸漏钢智能预报系统的研究[D];江南大学;2009年

6 朱文龙;基于遗传算法的BP神经网络在多目标优化中的应用研究[D];哈尔滨理工大学;2009年

，

本文编号：1597454