【作者】 陈昌；

【导师】 谢书明； 丁惜瀛；

【作者基本信息】 沈阳工业大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 转炉炼钢作为钢铁生产的重要环节，其主要目标是冶炼出温度和成分(主要指熔池碳含量)均合格的钢水。由于钢水的温度和碳含量不能连续检测，同时冶炼过程的边界条件变化频繁，这给冶炼过程的终点控制带来困难，在实际生产过程中，难以准确控制熔池碳、温而多次拉碳重吹，所以提高转炉炼钢终点命中率具有重要意义。由于转炉炼钢是一个非常复杂的多元多相高温物理化学过程，其机理的解析尚不透彻，输入输出间的非线性关系十分严重，常规建模方案始终不太理想。本文阐述了转炉冶炼终点控制技术的发展及概况，简要介绍了几种人工神经网络转炉炼钢终点预报模型。本文在原有静态控制方法的基础上，结合副枪检测信息，将智能控制应用到转炉炼钢过程的建模和控制中，提出转炉炼钢智能预报方法。并根据现场冶炼过程和数据，研究了转炉冶炼终点温度和碳含量的影响因素，确定了预报模型的输入变量，对常规的BP算法进行改进，利用Levenberg-Marquardt(LM)算法收敛速度最快，而且学习性能好的特点，分别建立了基于神经网络的终点碳、温预报模型。并与几种快速BP算法的特点及性能作了归纳和对比，结果表明LM算法具有比较高的预报精度。所提出的模型有严格的理论基础，经过了理论与试验的双重验证。模型选取某企业转炉车间现场60炉实际生产数据为样本，分别以影响转炉冶炼终点碳、温度的9个影响因素为输入变量，建立三层结构的BP神经网络模型，对终点碳、温度进行预报，并在此基础上确定补吹阶段需要的吹氧量和加入的冷却剂量，仿真研究结果表明了该方法效果较好，可以用于实际转炉炼钢过程。更多还原

【Abstract】 Basic Oxygen Furnace (BOF) Steelmaking is one of the most important parts of the iron & steel producing industry, the primary task of which is to provide the steel bath of which both the temperature and element (mainly the endpoint carbon content) hit the tipping aim slot at the steelmaking endpoint simultaneously. In actual BOF steelmaking process, steel bath element and temperature can’t be measured continuously and the operation conditions are very frequently, which makes it difficult to control the BOF endpoint precisely, and it often happens that operators have to re-smelt the steel bath due to the low control precision. So improving the control precision of BOF steelmaking process is very important. The most astringent fact is that the process of steelmaking is a synthesis of multi-substance, multi-phase, high temperature and attended by many physical chemistry reactions. The research on the mechanism of reactions is not very clear. And there are severe nonlinear and coupling relation between the inputs and outputs. So the effects of normal models are always not ideal.This paper has analyzed the development and actuality of control technology about the modem converter endpoint, introduced some artificial neural network models that are used to predict the endpoint of BOF steelmaking. In this paper, we combine the static control method based on the sublance information on BOF steelmaking process. And these methods are improved by introducing neural network. According to the process and data from spot, the paper has analyzed the factors for influence C and endpoint steel temperature in converter. The control variables for [%C], endpoint steel temperature prediction and control model were determined. We improve the BP algorithm and establish two prediction neural networks by the advantage of Levenberg-Marquardt (LM) algorithm that the convergence speed is the quickest and its performance is the most excellent for neural network. The .characteristic and performance of fast BP algorithms are generalized and contrasted; the results show that the model based on LM algorithm has higher precision.The proposed control model has a rigorous theoretic base and has been validated by the theory and experiment. We improve the BP algorithm and establish two prediction neural networks. The actual data of continuous 60 batches from a converter are chosen as example, 9 input variables that influence C and endpoint steel temperatures in converter are determined. We establish two three-layers BP prediction neural networks, and predict C and endpoint steel temperature content. The model of [%C] and endpoint steel temperature prediction have been established. On the basis of this, the method based on BP neural network was proposed so as to determine the blown oxygen and the added coolant during the re-blowing. Simulation results showed that the method is effective and can be used to practical BOF process.更多还原