【作者】 谭钢军；

【导师】 潘林强； 许进；

【作者基本信息】 华中科技大学 ， 系统分析与集成， 2013， 博士

【摘要】 高速线材水冷控制系统是高速线材自动化控制的重要组成部分,是实现轧线温度的智能控制,提高高线产品质量的关键。利用先进的计算机技术来完善水冷控制系统,可以提高高速线材产品质量,从而提高线材生产制造业的产品档次,实现传统钢铁企业的产业升级。在目前竞争激烈的高线材领域,高质量高规格的产品,可以开拓更广阔海内外的市场。因此,研究如何提高高速线材水冷控制系统的性能,特别是如何提高水冷控制系统的可靠性和控制精度具有很重要的现实意义。本文针对武汉钢铁集团公司大型轧钢厂在高速线材生产线中存在的水冷控制系统的可靠性差,轧线温度波动范围大等问题进行研究。应用智能计算理论及方法对上述工业控制系统进行系统辨识、建模以及优化,解决了高速线材生产线水冷控制系统中存在的温度控制精度波动范围大和温度控制系统可靠性差的问题,具体研究内容如下：针对高速线材生产线水冷控制系统中存在的温度控制精度波动范围大的问题,提出了一种具有学习自适应性且能逼近任意复杂函数的前馈多层感知器,通过建立基于前馈网络水冷控制系统的优化模型。通过大量采样SMS水冷控制系统的输入输出数据,在数据特征分析与预处理的基础上建立训练样本集,以高线水冷系统的温度控制环节为对象,以高线水冷系统的入口温度作为输入,精轧机出口温度作为输出,构建水冷系统的温度输入输出神经网络模型,实现了高线水冷控制系统的自适应温度控制。在仿真数据和实际数据上,实践表明,该系统投入应用后,轧线温度的控制精度有显著提高,轧线精轧机入口温度的波动范围由±60℃降低至±28℃,精轧机出口温度的波动范围由±30℃降低至±15℃。针对现有高速线材生产线水冷控制系统中实时性和精度难以平衡的问题,提出了两类基于膜算法的高线水冷控制系统优化模型。膜算法是一类受细胞结构和功能启发的分布式并行计算算法,是智能优化的一个新领域。近年来,膜算法在智能优化领域受到学者的广泛关注。本文利用膜算法优化高速线材生产线水冷控制系统,特别是对其BP神经网络的权值与阈值进行优化,在大量采集整个系统运行时通讯数据的基础上,通过膜算法训练,利用MATLAB工具箱构造的BP神经网络的权值和阈值,达到提高水冷控制系统的运行稳定性与控制精度的目的。在MATLAB软件仿真上,表明经膜算法优化后的水冷控制系统在满足工业实时性要求的同时,大幅度提高了输出精度。针对现有高速线材生产线的水冷控制系统中控制系统可靠性较差的问题,提出了一种基于贝叶斯概率网络的水冷控制系统故障判断及容错处理模型。通过贝叶斯网络自动判断采集数据的范围以及数据有效性,在线调整神经网络训练样本的数据集合,通过离线重新训练网络权值,并异步更新网络权值数据,从而在保证水冷闭环控制系统实时性的前提下,提高控制系统的自适应性与可靠性。理论表明,采用基于贝叶斯概率网络的水冷控制系统,大大提高了该系统运行的可靠性和控制精度。根据基于前馈网络水冷控制系统的优化模型,对武钢大型轧钢厂的高速线材水冷控制系统的控制软件进行优化,从实际数据结果表明,优化后的水冷控制系统提高了运行的可靠性和控制精度,实现了产业升级。更多还原

【Abstract】 The water cooling control system takes an important role in automatic control of high-speed wire rod mill. Since it realizes the intelligent control of temperature of rolling line, and ensures the quality of production, the water cooling control system is essential for the improvement of rolling speed. The use of advanced computer technology in the water cooling control system can enhance the monitoring, operation and automatic control level of the process of the wire rod production, thereby raising the efficiency and quality of the production. The above indicates that the improvement of the system of high-speed wire rod mills, especially the reliability and the precision of control is very meaningful.This dissertation has carried out in-depth research into the question of the low reliability and the high range of the temperature wave in the water cooling control system in high speed steel wire production line. The work has improved the reliability of the system and the precision of the temperature control.For the problem of the accuracy of temperature control in the water cooling control system in high speed steel wire production line, by introducing the feedforward multilayer perceptions that can approach complex function and has the ability of self-learning adaptive, we established an artificial neural network based optimization model in the water cooling control system and finally improved the accuracy of temperature control in water-cooling control system. Through a large number of input/output data samples of SMS water-cooling control system, we established the training sample set on the basis of data characteristics analysis and preprocessing, realized water-cooling control optimization system based on the Feed Forward Network (FFN). By the automatic judgment of the data gathering range and data validity with the guarantee of the real time, we established online adjustment method of the data sample collection for the neural network training in the water-cooled closed-loop control system, cooperate with offline training of network weights, and asynchronous update data network of weights. All these efforts results in improvement of the adaptability and reliability of the control system. The rolling line finishing mill inlet temperature range reduced from±60℃down to±28℃, the fluctuation range of the finishing mill exit temperature reduced from±15℃to±30℃.For the difficult to balance the accuracy and the real-time of temperature control in the water cooling control system, we utilized two types of membrane algorithm to improve artificial neural networks based optimization model in the water cooling control system. The membrane algorithm is a distributed parallel computing algorithm inspired by the cell structure and function, which is a new area of intelligent optimization. In recent years, membrane algorithm attracts the attention of scholars in the field of intelligent optimization. Experimental results showed that our methods not only meet the real time of the industrial requirements, but also greatly improve the accuracy temperature control. On the basis of data characteristics analysis and preprocessing, we introduced a method using the MATLAB toolbox to train the BP neural network weights and thresholds with the help of membrane algorithm. On site operation shows that the use of membrane algorithm greatly enhanced the stability and control accuracy of the water cooling control system.For the Robustness of the water cooling control system of high speed steel wire production line, we established the cooling system fault diagnosis and fault-tolerant model based on Bayesian probability network theory. Determine the range of data collection and data validation, Bayesian networks automatically adjust the neural network training sample data collection, offline re-training the network weights, and asynchronous updating of the weights of the network data, thus ensuring a water-cooled closed-loop control system in real time the premise, the self-adaptability and reliability of the network. On site operation shows that the use of Bayesian probability network optimization system greatly enhanced the reliability and control accuracy of the water cooling control system.According to the artificial neural network based optimization model in the water cooling control system, optimizing the control software of water cooling control system for Wuhan iron and steel company, from the results of the actual data, it is shown that optimized water cooling control system can improve the operational reliability and control accuracy and realized the industrial upgrading.更多还原