【作者】 徐喆；

【导师】 毛志忠；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2012， 博士

【摘要】 钢包精炼炉是炉外精炼的重要设备之一,在我国钢铁企业中有广泛的应用。钢包精炼可实现钢水温度与成分的精确调节、钢液净化、生产节奏调节等功能。因此,充分利用钢包精炼炉的精炼功能、优化各项精炼操作是提高钢材质量、降低生产成本、缩短冶炼周期、节约能源的有效手段。通过向钢水中加入合金物料以实现准确、快速地调节钢水中的合金成分是钢包精炼炉的重要作用之一,从而可确保钢包到达连铸台时的钢水合金成分合格,而合金料加入量设定值的准确程度主要受合金元素收得率预测精度的影响。在冶炼实践中,该参数的确定还停留在人为估计的阶段,其估计精度较低。.为此,本文主要针对提高合金元素收得率预测精度的问题,在深入分析钢水合金化冶炼机理的基础上,系统地进行了钢包精炼过程元素收得率预测模型建模方法的研究。本文的主要研究工作归纳如下：1.为了确定收得率预测模型的输入变量,首先从合金化的基本原理出发,简要分析了影响合金元素收得率的主要因素。但由于检测手段的限制,一些影响因素无法直接获得。为解决这个问题,依据冶炼机理,理清了影响因素与可测变量间的对应关系,以便使用可测变量间接反映收得率。最终确定了一组变量作为收得率预测模型的输入变量。通过数据仿真,从统计的角度验证了这组变量的有效性,为建立收得率预测模型奠定了基础。2.针对一些钢种冶炼数据量少的特点,提出了基于规则融合的模糊建模方法用以建立收得率预测模型。该方法使用冶炼经验知识弥补训练数据覆盖范围不足的问题,将经验知识以TSK (Takagi-Sugeno-Kang)型模糊规则的形式引入模糊模型中。提出了模糊规则融合方法,在建模的结构辨识过程中,将经验规则与从数据中提取的数据规则进行结合,用以确定初始的模糊规则。在参数辨识阶段,在原梯度下降方法目标函数中引入了经验知识评价参数,用以平衡样本数据与经验知识对模型的贡献程度。数值仿真与工程实例验证结果表明,该方法可以有效利用经验知识与样本数据,使预报结果更可靠、更精确。3.针对常炼钢种数据量多的特点,提出了基于工况划分的多模型建模方法用来建立合金元素的收得率预测模型。工况的准确划分是该方法的基础,为此,根据数据输入变量服从近似正态分布的特点,提出了基于超椭球的数据划分方法,该划分方法按数据分布的疏密程度将数据集划分为稀疏子集与密集子集两部分,然后分别在两个数据集上建立独立的预测模型。稀疏子集数据量少且分布稀疏,为使用此数据集建模,提出了基于改进遗传算法的聚类变量构建方法,用于聚类变量的简约以便工况识别,在获得简约的输入变量后,对其采用模糊ISODATA聚类方法按工况进行模糊划分。在划分后的数据集上利用原输入变量建立基于分类回归树的多模型预测模型对隶属于稀疏数据区域的炉次进行合金元素收得率的预报。对于密集数据集,其特点为数据量大且分布集中,为更有效地对其建模,采用了在线的即时学习方法,并根据输入数据的分布特点对该方法的相似数据评价准则进行了改进,以便使选取的参考数据可以更全面地反映当前待预测炉次的工况。该方法的特点为只使用与当前炉次同工况的数据建模,从而也降低了引入异常数据的风险。最后,通过仿真实验的结果可以看出,基于工况划分的多模型方法较其它单一模型的建模方法有更高的预测精度。4.以某钢厂精炼工段作为具体研究对象,采用了由执行层、控制层和优化层构成的三层体系结构,在基础自动化系统的支持下,以上述理论研究为基础,设计并开发了钢包精炼钢水合金元素成分优化控制操作系统软件。实现了合金元素收得率预测以及合金优化配料计算。将软件实际应用于三个钢种的精炼操作中,取得了较好的合金成分控制效果与经济效益。更多还原

【Abstract】 As one of the important equipment in external refining, ladle furnace has been widely applied in the iron and steel enterprise in China. Through ladle furnace refining, functions such as the precise regulation for the temperature and composition of the molten steel, purification of molten steel, regulation for the rhythm of production and etc. can be realized. The effective approaches for improving the steel quality, reducing the production cost, shortening the smelting cycle and saving energy is to take full advantage of the refining function of the ladle furnace and to optimize various refining operations. Accurate and quickly adjustment of the content of alloying compositions in molten steel by adding alloy materials is one of the important role of ladle furnace refining. It can ensure that the content of alloying compositions of the molten steel is qualified when the molten steel arriving at ladle turret. While, the accuracy of the setting values of the amount of alloy materials to be added are mainly determined by the prediction accuracy of alloying element yield rates. In smelting practice, the determination of this parameter still stays in the stage of human estimation with lower accuracy. Therefore, targeting at improving the estimation precision of the alloying element yield rate, and on the basis of further analyzing the smelting mechanism for molten steel alloying, the study on the modeling for element yield rate estimation during the ladle refining process has been conducted. The major research work is summarized as follows:1. In order to determine the input variables of the prediction model of alloying element yield rate, firstly, the major factors influencing the yield rate of the alloying element have been analyzed simply by starting from the basic principles of alloying. However, due to the limitations in detection means, some influencing factors cannot be obtained directly. In order to solve this problem, the corresponding relationship between influencing factors and measurable variables has been figured out according to the smelting mechanism, which will be convenient for reflecting the yield indirectly with measurable variables. Finally, a group of variables have been determined as the input variables of the yield rate prediction model. Through data simulation, the effectiveness of variables in this group has been testified from the perspective of statistics, which has laid a foundation for the establishment of yield rate prediction model.2. Fuzzy modeling method based on the rule fusion has been proposed to establish yield rate prediction model of alloying element by targeting at the characteristic that there is little data for some steel grades. In this method, the smelting experiential knowledge has been adopted to make up for the shortages in the covering scope of training data, and experiential knowledge has been introduced into fuzzy model in the form of TSK (Takagi-Sugeno-Kang) fuzzy rules. The fuzzy rule fusion method has been proposed, and in the structure identification process of modeling, the experiential knowledge has been combined to the data rule extracted from the data so as to determine the initial fuzzy rules. In the parameter identification stage, the experiential knowledge evaluation parameter has been introduced into the objective function of original gradient descent method to balance the contribution from sample data and experiential knowledge to prediction model. The testing results of data simulation and engineering project show that, in this method, the experiential knowledge and sample data can be used effectively to make the prediction results much more reliable and precise.3. In accordance with the substantial data for common steel grades, multi-model modeling method based on working condition division has been proposed to establish the yield rate prediction model for alloying element. The accurate division of the working condition is the basis of this method. Therefore, according to the characteristic that the input variables obey normal distribution approximately, data division method based on hyper-ellipsoid has been proposed, which divides the data into sparse subset and dense subset according to the density of data distribution, and then independent prediction models have been established on the two data sets respectively. There is little data distributing sparsely in the sparse subset, and as a result, clustering variable construction method based on the improved genetic algorithm has been proposed to employ this data set for modeling, in which the clustering variable simplicity is convenient for the identification of working condition. After the acquisition of simplified input variables, the working condition will be divided fuzzily with fuzzy ISODATA clustering method. Original input variables belong to the divided sparse data sets are adopted to establish the multi-model prediction model based on the classification and regression tree for the prediction of alloying element yield rates of the heat affiliated to the sparse data region. The dense data set is characterized with substantial data and concentrated distribution, to modeling in a more effectively way, the online just-in-time learning method has been adopted, and the evaluation criterion of data similarity has been improved according to the distribution characteristic of the input data, so that the referential data selected can reflect the current working condition of the current heat comprehensively and conveniently. The characteristic of this method lies in that only the data concerning the current heat working condition are used for modeling, thus the risks of introducing abnormal data have been reduced. Finally, it can be seen from the results of simulation experiments that, the multi-model modeling method based on the division of working condition has higher prediction precision than other modeling methods with single model.4. Taking the refining section in some steel mill as the specific research object, the three-layer architecture composed by execution layer, control layer and optimization layer has been adopted. Under the support of basic automation system and on the basis of the theory stated above, the operating system software for the optimal control of alloying compositions in ladle furnace refining has been designed and developed. And the estimation calculation for the yield rates of alloying elements and optimization of alloying ingredient has been achieved. When applying the software into the refining operation for three steel grades, better control effect for alloying composition and economic benefit have been achieved.更多还原