【作者】 赵永磊；

【导师】 郑德忠；

【作者基本信息】 燕山大学 ， 测试计量技术及仪器， 2010， 博士

【摘要】 板带材在国民经济各部门中具有广泛而重要的应用。板带材的质量指标之一是板形。由于板带材使用部门对板形精度的要求越来越高,使板形控制成为现代高精度板带轧机的重要的技术发展方向,其中板形在线预测与板形先进控制方法与策略的研究则是现代板带轧制中的关键技术和国际前沿研究课题。本文以广义预测与人工智能理论为基础,选择具有理论和工程实际意义的冷轧机板形在线智能预测与液压弯辊广义预测控制为研究课题,进行了深入的理论研究与工业应用研究,主要工作如下:板形预测模型是板形控制系统设计的重要基础,无论是板形控制系统中的调节机构控制特性分析,还是在线实时控制,都需要精确的板形预报模型。针对目前基于机理模型的板形预测模型在板形预测中效果不好的现状,为提升板形预测模型的快速性和准确性,本文以生产实测数据为基础,建立了基于模糊分类的分布式神经网络板形在线预测模型。将对角递归神经网络及模糊分类技术引入到了板形预测中,并给出一种用于板形在线预测模型的校正算法,使预测模型可以适应变化的过程特性以获得较好地预报结果,提高了板形预测的准确性和系统的实时性。该预测模型克服了机理模型中的反复迭代、计算时间长以及多层前馈神经网络易将动态建模变成静态建模问题的缺点,探索了一种非解析原理的板形建模方法。针对液压弯辊控制具有非线性、时变性及不确定性等特性,及其对抗干扰性的要求,在分析其控制机理的基础上,提出了一种基于直接广义预测控制的液压弯辊控制方案并将其应用于带钢板形控制中,以提高带钢的成才率,充分发挥液压弯辊力对板形的调整作用,改善轧机系统的动态特性。针对传统广义预测控制算法由于存在反复求逆而导致计算量过大的问题,提出了一种新的简化迭代优化算法,利用共轭梯度法对滚动优化性能指标求解极值来计算最优控制律,省去了广义预测控制求逆计算过程,以解决广义预测控制算法计算量大而影响实时性的问题。研究了广义预测控制的模型反馈校正方法,将递推极大似然法和遗忘因子递推最小二乘法结合起来,给出了一种改进的递推极大似然参数估计算法。该算法可及时修正模型误差、提高控制精度并保证控制效果。解决了控制参数与噪声干扰紧耦合时参数估计变慢的问题,可有效地抑制系统时滞、设备性能时变对系统稳定性的影响。论文最后对板形在线预测模型和广义预测控制在板形测控系统中的应用进行了进一步的讨论和设想,预示了其广阔的发展空间和应用前景。更多还原

【Abstract】 In every department of national economy, plate and strip has extensive and important application. One of the quality indexes is flatness. The precision of plate and strip acquired by department is becoming more and higher. As a result, flatness control becomes the key technique and important develop direction for modern times high precise rolling mill. As far as the technique is concerned, theoretical basis and key scientific problem is the online flatness forecasting and advanced control method. In this study, author chooses flatness intelligent forecast and hydraulic roll-bending control as research object. The system of flatness measurement and control has been researched based on Generalized Predictive Control (GPC) and Artificial Neural Network (ANN). Main works are as follows:Flatness forecasting model is important to design flatness control system and precise flatness forecasting model is needed either in analyzing the control characteristic of the machines adjusted or in controlling online. In order to build a more fast and precise model of flatness forecast, an online distributed neural network flatness forecasting model is built based on product data. Diagonal recurrent neural network and fuzzy classification technique are introduced in the flatness forecasting model. A Correction Algorithm that used to forecasting model was set up, which cause model can adapt process changes and can obtain better forecasting result. The forecasting model overcomes a lot of defects, such as iterative operation, time-consuming operation, easily making dynamic model static model in many lay ahead feedback network, exploring a new non-parse method of building flatness model and resolving many problems in building complex system model.Due to non-linearity, time-variation and non-determinacy of hydraulic roll-bending control, as well the anti-interference request, a hydraulic roll-bending control scheme has been put forward based on direct GPC after analysis of system mathematical models, which is used to raise product ratio, make full use of hydraulic force and improve the dynamic performance of rolling mill system. GPC has been studied. According to the problem of big calculation amount caused by heavy and complicated matrix inversion, a new simplified iterative optimization algorithm has been presented to solve the GPC real-time problems caused by the amount calculation, which eliminate the matrix inversion of traditional GPC.Model feedback correction algorithm of GPC has Benn studied. Combining the RML and forgetting factor RLS, An improved RML parameter estimation method has been given. This method can modify model error in time to improve control precision and guarantee control effect. It can solve the problem that parameters estimation will become slow when the process parameters and noise signal are tightly coupled. The effect of time varying of systematic time lag and equipments capability for systematic stability and robustness has been effectively restrained.At last ,the paper further we discusses and assumes the application foreground of flatness online forecasting and GPC in flatness measure and control system, and shows their board development space and apllication foreground.更多还原