【作者】 李帅；

【导师】 徐林； 王向华；

【作者基本信息】 东北大学 ， 控制工程（专业学位）， 2014， 硕士

【摘要】 转炉炼钢是当今世界上最主要的炼钢方法,我国一些先进的钢铁企业的转炉已采用动态控制技术。转炉炼钢是一种极其复杂的工业过程,影响终点温度和碳含量的因素很多,由于炉内的温度过高,对终点温度和碳的含量不能及时、准确地测量,因此建立精确的温度和碳的预报模型就显得十分重要。针对这个问题本文建立了转炉炼钢神经网络终点预测模型,根据预报结果可以对补吹氧气量及冷却剂加入量进行合理调整,从而提高终点命中率,以提高转炉炼钢产量和质量,减少能源消耗,降低炼钢成本。本文在鞍钢信息产业公司的冶金全流程仿真项目的基础上进行深入的理论研究和延伸。主要研究工作如下：由于转炉炼钢神经网络模型的核心是其终点预测模型,由于其工艺复杂,影响因素多,首先利用粗糙集属性约简的方法,对转炉输入属性进行约简,再结合实际的现场数据训练神经网络,从而得到更好的预测模型。RBF神经网络与BP神经网络相比学习时间短,具有很好的非线性预测效果。但是由于转炉炼钢工艺复杂,数据繁多,因此本文对传统的RBF神经网络模型进行了改进,利用蚁群聚类的算法确定基函数中心和隐含层节点数目。为了解决训练样本少、预测结果不准确的问题,引入灰色GM(1,N)预测模型改进RBF神经网络,将改进后的模型进行整合得到一种复合预测模型——灰色RBF神经网络预测模型。本文运用各神经网络预测模型和灰色RBF神经网络预测模型对钢厂的实际冶炼数据进行的仿真,改进的预测模型仿真效果结果明显要好于其它模型,说明本文所提出的方法具有可行性。更多还原

【Abstract】 BOF is the main steelmaking methods in the modern world, some advanced iron and steel enterprises in our country has adopted dynamic control technology. BOF is an extremely complex industrial processes, many factors affect the endpoint temperature and carbon content, because of the furnace temperature is too high, the endpoint temperature and carbon content can’t timely and accurate measurements, thus establish accurate forecasting model of temperature and carbon is very important. Aimed at this problem,a BOF endpoint prediction model of neural networks is established in this thesis, according to the forecast results can fill of blowing oxygen and make reasonable adjustments for the amount of coolant, thus improve the finish shooting, in order to improve the yield and quality of converter steel-making, reduce energy consumption, reduce the cost of steelmaking.Based on the actual project of the whole process simulation project in AnSteel information industry company,this thesis dose the in-depth theoretical research and extension, the main work is as follows:Because of the end point prediction model is the core of the BOF neural network model, due to the complexity of its process and many of affecting factors, first of atl, using rough set attribute reduction method, reduce the converter input attribute, then combined the actual field data training neural networks, so as to get a better prediction model.Compared with BP neural network, RBF neural network learning time is short, and has good nonlinear prediction effect. However, due to steelmaking complex process, the variety of data, therefore, in this thesis, the traditional RBF neural network model is improved, the center of ant clustering algorithm is used to determine the basis function and the number of hidden layer nodes. In order to solve the prediction of less training samples, the problem of inaccurate, introduce the gray GM (1,N) model improved RBF neural network, the improved model integrated into a composite forecast model, gray RBF neural network prediction model.This article use various neural network prediction models and gray RBF neural network prediction model, use the actual smelting data to simulate, the improved forecasting model simulation result is obviously better than the other models, it shows that the proposed method is feasible.更多还原