基于神经网络的高炉铁水硅含量建模

发布时间：2019-04-01 18:49

【摘要】：钢铁冶金行业作为我国基础性原材料的源头产业,一直是我国国民经济的支柱产业。高炉炼铁作为钢铁制造主体的上游工序,是钢铁工业的重要构成环节,对行业整体的降耗节能与发展都起到至关重要的作用。炉况是否顺行直接关系到整个炼铁过程的节能减排,而高炉炉温则是鉴别高炉炉况,保证高炉炉况顺行的一个重要指标。铁水硅含量作为炉热状态的表征指标,建立其可靠的预报模型以指导高炉炼铁人员进行炉温控制不仅具有重要的理论研究价值,而且对关键的生产实践具有重要的指导意义。然而,高炉冶炼的过程异常复杂,在建模方面的主要难点表现在运行机制时常伴有非线性、大噪声、分布参数等特征。而神经网络作为自学习网络的一种,能够比较很好地解决非线性、大噪声的问题。到目前为止,神经网络已经在模式识别、预测控制、函数逼近等领域得到广泛的应用。本文针对铁水硅含量建模这一关键问题,利用神经网络方法进行建模研究,具体研究内容如下。首先,针对高炉冶炼过程中的相关数据进行分析处理。利用专家经验和数据统计分析相结合的方法确定铁水硅含量的相关输入变量,进而利用多元线回归法确定滞后时间序列的相关系数。所有数据均进行了一次和二次处理,且进行了归一化处理。其次,针对BP神经网络算法容易陷入局部最优的缺点,提出了改进方案。仿真数据按照所提出的方案进行了分析处理,其对比仿真验证了改进BP神经网络在铁水硅含量建模上的优势。最后,提出了新的基于正则化方法的极值学习机算法,本算法可以克服常规极限学习机的缺点。进而利用该算法对对高炉炉温进行建模。实验仿真验证了本文算法的可行性,并且证明了正则化的极值学习机优于BP神经网络。
[Abstract]:As the source industry of basic raw materials in China, iron and steel metallurgy industry has always been the pillar industry of our national economy. Blast furnace ironmaking, as the upstream process of iron and steel manufacturing, is an important part of iron and steel industry, and plays an important role in reducing consumption and energy saving and development of the industry as a whole. Whether the furnace condition runs smoothly or not is directly related to the energy saving and emission reduction of the whole ironmaking process, and the blast furnace temperature is an important index to distinguish the blast furnace condition and ensure the smooth operation of the blast furnace condition. The silicon content of molten iron is regarded as an indicator of furnace thermal state. The establishment of a reliable prediction model to guide blast furnace ironmakers to control furnace temperature not only has important theoretical research value, but also has important guiding significance for key production practice. However, the process of blast furnace smelting is very complex, and the main difficulty in modeling is that the operation mechanism is often accompanied by nonlinear, large noise, distributed parameters and so on. As a kind of self-learning network, neural network can solve the problem of non-linear and large noise. So far, neural networks have been widely used in many fields such as pattern recognition, predictive control, function approximation and so on. In this paper, a neural network method is used to model the silicon content in hot metal, and the main contents are as follows. First of all, the relevant data in blast furnace smelting process are analyzed and processed. The correlation variables of silicon content in hot metal are determined by the method of expert experience and statistical analysis of data, and the correlation coefficient of delay time series is determined by multiple linear regression method. All data were processed once and twice, and normalized. Secondly, aiming at the disadvantage that BP neural network algorithm is easy to fall into local optimization, an improved scheme is proposed. The simulation data are analyzed and processed according to the proposed scheme, and the advantages of the improved BP neural network in the modeling of silicon content in hot metal are verified by the comparison of the simulation results. Finally, a new extreme learning machine algorithm based on regularization method is proposed, which can overcome the shortcomings of conventional limit learning machine. Then this algorithm is used to model the blast furnace temperature. Experimental results show that the proposed algorithm is feasible and the regularized extreme learning machine is superior to BP neural network.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TF325.6;TP183

【相似文献】