高炉料面形状检测与预测方法研究

发布时间：2018-04-11 17:12

本文选题：高炉料面 + 群智能算法　；参考：《北京科技大学》2017年博士论文

【摘要】：钢铁工业是我国国民经济长期的支柱性产业,是发展其他产业的重要基础,在社会发展和经济建设中发挥着举足轻重的作用。高炉是钢铁生产过程中的关键设备,关系到行业的钢铁产量、能源消耗和环境污染。维持高炉的长期稳定、高效运行,是钢铁行业追求的共同目标。高炉内部料面分布是影响高炉炉况的重要因素之一,对于维持高炉煤气流合理分布、增加料层透气性和高炉优化操作有重要的作用。本文针对高炉的密闭性,不能准确、直观检测料面状态的问题,运用高炉雷达检测技术、高炉料面形成机理、群智能算法、智能计算等技术进行“高炉料面形状检测与预测方法”的课题研究。主要从雷达料面检测传感器安装位置优化部署、高炉料面检测、料面下降速度预测以及料面异常诊断四个角度深入研究,对高炉的长期稳定运行和节能减排具有一定的指导和应用价值。本文的主要研究内容及创新点包括以下4个方面：(1)提出一种高炉雷达传感器安装位置的优化方法,方法可以兼顾已安装的其他类型传感器,实现信息统一,避免冗余。同时,保证在使用最少雷达数量的前提下,实现雷达对炉喉料面的完全覆盖以及料面关键点的K-重覆盖。优化方法一方面根据炉喉半径与雷达覆盖直径的关系,建立料面的环形区域,减少雷达安装数量,加快优化速度：另一方面,根据料面形状特征分析,给出传感器优化部署的评价函数,利用改进的人工鱼群智能算法求解优化问题。最后,分别利用标准测试函数和现场实际数据验证方法的有效性。(2)提出一种重构高炉料面形状的模型框架,模型基于布料规律和多点雷达数据实现对高炉料面的重构。首先提出料面形状描述方程,根据力学原理,求解料面堆角和堆尖位置,建立基于布料规律的料面方程;其次,提取雷达传感器采集的料面实时高度数据,引入多源信息融合的思想,对基于布料规律的料面方程进行修正。最后,基于炉料体积约束原则,采用迭代计算确定料面方程。经过与其他算法的测试比较,提出的重构模型具有较高的精度,料面方程更加合理。(3)提出高炉料面下降速度预测模型,首先,对表征高炉料面变化的雷达时间序列运用C-C算法计算时间延时和嵌入维数,重构相空间,利用小数据量法计算最大Lyapunov指数,证明料面高度变化具有混沌性,为混沌预测方法的应用建立了理论基础；其次,分别采用极限学习机和在线惯序极限学习机建立高炉料面下降速度的离线和在线预测模型；最后,利用实际生产数据进行测试,并与同类算法比较,结果表明,预测算法在预测精度和速度上有良好的表现,说明混沌方法在预测料面下降速度方面的应用是可行的。(4)提出高炉料面异常的不平衡样本分类预测模型,模型针对高炉料面异常样本数量少,造成训练样本类别分布不平衡的特点,对旋转森林集成算法的样本选取和建立基分类器两个阶段分别进行优化,建立料面异常分类预测模型。用标准测试集进行测试,并与其他集成算法比较,改进算法在总体精度和对少数类别分类精度都有所提高；利用实际生产数据建立分类预测模型,对料面异常样本具有较高的分类精度。
[Abstract]:The iron and steel industry is China's long-term national economic pillar industry, is an important foundation for the development of other industries, plays an important role in the social development and economic construction. The blast furnace is the key equipment in steel production process, related to the industry of steel production, energy consumption and environmental pollution. To maintain long-term stability of blast furnace the efficient operation of the steel industry is the pursuit of common goals. The burden distribution of blast furnace is one of the important factors that affect the internal state of blast furnace, blast furnace gas flow to maintain reasonable distribution, increase the permeability of blast furnace operation and optimization has an important role. In this paper the sealing property of the blast furnace can not accurately detect the state of charge, visual problems the use of radar detection technology of blast furnace, the formation mechanism of blast furnace, swarm intelligence algorithm, intelligent computing technology for blast furnace shape detection and prediction method of the research. Mainly from the radar level detection sensor location optimization deployment, blast furnace detection, four aspects of in-depth study of abnormal diagnosis rate prediction and burden descent level, long-term stable operation of blast furnace and energy saving has certain guiding significance and application value. The main research contents and innovations of this paper include the following 4 aspects: (1) put forward the optimization method of blast furnace radar sensor installation location, method can take into account other types of sensors have been installed, to achieve unified information, to avoid redundancy. At the same time, ensure the quantity of at least under the radar, radar on the surface and throat completely cover the surface of key point K- optimization method according to the coverage. The relationship between the furnace throat radius and the diameter of the annular region of radar coverage, establish level, reduce the number of radar installation, to speed up the optimization speed. On the other hand, according to the surface The shape feature analysis, optimal deployment evaluation function is given, using the improved artificial fish swarm algorithm for solving the optimization problem. Finally, the validity of the standard test functions and actual data validation method respectively. (2) proposed a model frame reconstruction of blast furnace shape, reconstruction of the implementation of the rules and distribution of blast furnace burden multi radar data. Based on the first proposed surface shape description equation, according to the mechanics principle, solving the charge level angle and pile tip position, establish the equations based on the laws of surface cloth; secondly, the real-time surface height data extraction of radar sensor acquisition, introducing the idea of multi-source information fusion, the surface equation based on modified distributing law finally, the charge volume constraint based on the principle of using iterative calculation to determine the surface equation. After testing and comparison with other algorithms, the proposed model is reconstructed High precision, surface equation is more reasonable. (3) the blast furnace burden descent speed prediction model, firstly, the characterization of blast furnace burden change radar time sequences by C-C algorithm to calculate the time delay and embedding dimension of phase space reconstruction, using a small amount of data to calculate the maximum Lyapunov index, that burden height change is chaos, application chaos prediction method established a theoretical foundation; secondly, using extreme learning machine and used online offline and online order limit learning machine to establish prediction model of blast furnace burden rate of decline; finally, with the actual production data were tested, and compared with other similar algorithms, the results show that the prediction algorithm has good performance in prediction accuracy and speed, speed down that chaos method in the prediction of application level is feasible. (4) the blast furnace burden imbalance abnormal sample classification prediction model Type, model number for blast furnace abnormal samples, resulting in uneven distribution of the training samples, the rotation forest integration algorithm and sample selection of base classifier is established in two phases optimization, establish surface anomaly classification prediction model was tested using the standard test set, and compare with other integrated algorithm, improved algorithm the overall accuracy of minority class and classification accuracy are improved; the classification prediction model is established by using the actual production data, has higher classification accuracy on the level of abnormal samples.

【学位授予单位】：北京科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TF54;TP18

【相似文献】