【作者】 赵敏；

【导师】 刘祥官；

【作者基本信息】 浙江大学 ， 运筹学与控制论， 2008， 博士

【摘要】 高炉炼铁是钢铁工业的上游主体工序,作为国民经济支柱产业的重要组成部分,它对整个钢铁工业的发展与节能降耗都起重要的作用。本文以内蒙古包钢6#高炉在线采集的冶炼过程数据为基础,针对炼铁生产中工长关注的实际问题,主要从炉温变动的复杂性、炉温预测这两个角度深入研究高炉冶炼过程的复杂性机理及其预测,具有一定的理论意义和应用价值。高炉炼铁过程的炉温预测数学模型研究,既是炼铁自动化中的难题,也是实际生产中工长和厂长都十分关注的重要课题。分析炉温波动并对炉温进行准确预报,将有助于提高工长的操作水平,从而达到提高利用系数和降低焦比的目的。论文的第二章对表征铁水质量和高炉炉温的含硅量[Si]时间序列分别运用替代数据法和信息冗余图像法做定量和定性的非线性检验,证明[Si]序列存在较强的非线性相关,为非线性预测方法的应用建立了理论基础。并从平稳性和序列演化的确定性角度进一步研究铁水[Si]序列的波动特征,统计逆序检验的结果表明:[Si]序列方差具有显著的非平稳特征,而相空间重构后的目标方差检验初步显示:铁水硅含量序列的可预测性介于确定性系统和混沌Lorenz系统产生的时间序列之间。在此基础上,论文第三章结合包钢6#高炉在线采集的铁水含硅量数据,应用经验模式分解、支持向量机、Volterra级数、Hopfield网络等方法从非平稳和非线性的角度对[Si]序列做单变量拟合和预报。考虑到高炉冶炼过程是高炉主控制室指挥下的多工序复合系统,对铁水含硅量有影响的因素错综复杂,因此,仅仅考虑单一硅序列对高炉铁水含硅量做预报是不够的。根据包钢6#高炉数据采集的实际情况,论文第四章选取了风量、喷煤、料速等14个变量,通过灰色关联度和粗糙集方法深入分析变量与铁水含硅量之间的非线性相关度及这些变量之间的内部关联,对冗余信息做知识约简,确定了起主导作用的6个变量。依照所确定的变量,第五章针对现有模型不考虑噪声污染、不够稳健、无法确定输入变量与[Si]之间具体函数关系的缺点,分别运用集员辨识理论和遗传程序设计方法对包钢6#高炉连续100炉铁水含硅量[Si]做多变量时序预测,两种算法得出的命中率分别为85%和83%,具有一定的实际应用指导价值。第六章对全文的研究内容以及创新点做了归纳,并对课题的后续研究做了展望。更多还原

【Abstract】 Blast Furnace(BF) ironmaking,which is the main working procedure of the metallurgical industry,is the pillar of the national economy,and is very important for the development of iron & steel industry and economizing energy consumption. Taking Blast Furnace smelting data obtained form No.6 BF at Baotou Steel as the foundation,the complexity of silicon content fluctuation and prediction of furnace temperature were studied in detail.The metallurgical figures pay extremely attention to the development of the mathematical prediction model of furnace temperature,because analyzing the behaviors of the furnace temperature fluctuation and accurately forecasting the temperature is the key,with which,BF operators can control ironmaking process well, the utilization coefficient of BF will increase and ratio of coke burden will decrease, Hot metal silicon content([Si]) is an important index in BF ironmaking process.Not only is silicon content a significant quality variable,it also reflects the thermal state of BF and can be used to represent the furnace temperature.As[Si]time series of No.6 BF at Baotou Steel to be sample space,the quantitative surrogate data technique and qualitative method based on information-theoretic functionals-redundancies(linear and nonlinear forms) are used to test nonlinearity in time series.The results show that there is intrinsic nonlinearity in[Si]time series and provide firm rationale for the nonlinear prediction and control of furnace temperature.To further explore the fluctuation characteristic of silicon content series,the stationarity test based on the number of reverse order and the DVV(Delay Vector Variance) predictability examination are implemented.The conclusion is:although the variance of[Si]time series is nonstationary,there still exists deterministic components in ironmaking process,the predictability of[Si]series is between time series generated from Lorenz chaotic system and deterministic system.On the basis of identification of nonlinear and nonstationary characteristics,Chapter 3 proposes two predictive models using empirical mode decomposition technique,support vector machine,Volterra theory and Hopfield network method.Recognizing that Blast Furnace ironmaking is a complex system consisting of multi-procedure,a range of factors is involved in influencing the silicon content in hot metal.The prediction models are insufficient if we take no account of the influences of some operational and controlling factors.In accordance to data acquisition capability of No.6 BF at Baotou Steel,fourteen variables,such as blast volume FL. pulverized coal injection PM and material velocity LS,were selected.Employing the grey incidence analysis and rough set method,Chapter 4 provides precious insight into the analysis of relationships among these variables and silicon content.Then six crucial factors were finalized.In view of the extant models’ shortcomings,such as instability causing by noise pollution,uncertain functional relation between variables and silicon content,etc,Chapter 5 made multivariate time series predictions respectively by set membership theory and genetic programming method according to the selected factors.The hit rates of silicon content are 85%and 83%.These algorithms come up to the industry index and they are valuable for practical application.Finally,Chapter 6 gives the conclusion and theoretical innovations in this paper,issues for further research are also investigated.更多还原