【作者】 贺诗波；

【导师】 刘祥官；

【作者基本信息】 浙江大学 ， 运筹学与控制论， 2008， 硕士

【摘要】 钢铁是现代人类社会使用的最广泛和最重要的材料之一,是国民经济持续发展的基础。高炉炼铁是钢铁工业的上游主体工序,它的发展直接关系到后续工序的发展,且对钢铁工业的节能降耗起着重要作用。然而,由于高炉是一个极其复杂的系统,其运行机制往往具有非线性、时滞、高维、大噪声、分布参数等特性,至今对其运行机理尚未能深入了解。高炉冶炼过程的这种复杂性,增加了其过程控制自动化的难度。一直以来,冶金自动化领域都没有攻克这个技术难题。在非平稳炉况下对高炉炉温进行预测和控制,则是这个难题的重要组成部分,更是当代冶金科技发展的前沿课题。本文以包钢6#高炉(2500m~3)在线采集的数据为研究对象,针对现有炉温预测模型的种种不足,在科学分析的基础上,第一次提出在高炉系统中用自组织数据挖掘方法进行系统分析和建模。这种方法集合神经网络、遗传算法和回归分析的相应优点,能在给定系统输入和模型选择准则(外准则)后自动进行模型的筛选,并且能充分考虑变量之间的相互影响。经过在高炉现场的大量调研,并结合实际经验和相关理论给出了影响系统的输入变量、中间网络的传递函数和模型选择准则。应用自组织数据挖掘的参数GMDH算法,得到了一个自动生成的高炉炉温预测控制模型。实际应用中,这个模型有着较好的预测命中率。并且由于模型中考虑到变量之间的相互影响,此模型也可以作为高炉系统分析的基础。接着用自组织数据挖掘的非参数方法对系统进行了研究,得出了非参数方法在高炉系统应用的优缺点。因此,引入自组织数据挖掘算法对于解析高炉系统运行的机理提供了一个非常有力的工具。本文以高炉冶炼现场的实际需要为根本出发点,以数学理论、高炉冶炼过程的工艺机理为基础,对高炉运行过程进行了详细的分析。实践表明,本文建立的数学模型对指导高炉炼铁生产实践,进一步研究高炉系统具有重要的理论价值。更多还原

【Abstract】 Steel is one of the most important and widely used materials in modern society, it is the basis of sustainable development of national economy. As the upper procedure in the iron and steel industry, Blast Furnace (BF) iron-making has great influence on the succeeding process of steel making. It is also a prime factor for the energy saving and consumption reduction. However, since BF iron-making process is highly complicated, whose operating mechanism is characteristic of nonlinearity, time lag, high dimension, big noise and distribution parameter etc, its mechanism is still unknown. The complexity of the whole process makes it difficult for automatic control. The most difficult part of automatic control for BF iron-making is to make accurate prediction of silicon content in hot metal under instable status and use the predicted information to control the iron-making process, which is also the frontier of research in the field of metallurgical automation.The current work uses data collected from Blast Furnace No. 6 in Baotou Steel (with an inner volume of 2500m~3) as sample space. A profound analysis of the existing predictive methods for silicon content is exercised. To overcome the deficiencies of existing methods, a self-organizing data mining method is proposed. This method combines the merits of neural networks, genetic algorithm and regression analysis and can choose the model automatically after giving the inputs of system and model selection criterion. And its biggest merit is that it takes interaction between the influencing factors into consideration. Through in-depth investigation, the input variables that carry most information of the system, as well as transfer function of middle network and model selection criterion are introduced. The GMDH algorithm is then used to construct a model for prediction of silicon content in hot metal and good results are obtained. The interaction between input variables is helpful for analysis of the system. To further explore the algorithm, a non-parameter self-organizing data mining method is used and the merits and shortcomings of it are gained. The self-organizing data mining algorithm offers an effective tool for the understanding of the mechanism of BF iron-making. Based on mathematical theories and the technical mechanism of iron-making process in BF, an in-depth analysis of blast furnace iron-making is exercised and the self-organizing algorithm is implemented to predict silicon content in hot metal. Simulation results show that the mathematical models in this paper are effective and that they are helpful for further research of BF system.更多还原