基于多率采样的焦炉火道温度软测量集成模型

【作者】 陈泰任；

【导师】 曹卫华； 吴敏；

【作者基本信息】 中南大学 ， 控制科学与工程， 2008， 硕士

【摘要】 焦炭是钢铁等行业的重要生产原料,广泛的应用于各行各业。焦炉加热燃烧过程是炼焦过程中重要的流程,火道温度直接影响到焦炭质量和能源消耗。然而由于成本等原因,往往难以实现实时测取火道温度;而人工测温,也存在周期太长,迟滞的缺点。本文针对焦炉火道温度不易测量,过程参数具有非线性、时变等特点,探究基于多率采样的焦炉火道温度软测量集成模型,在焦炉燃烧过程中的应用,实现实时测取火道温度的目标。在细致分析本课题的研究背景和意义的基础上,本文采用软测量集成建模的思想来分析焦炉燃烧过程。从炼焦工艺出发,通过深入研究焦炉加热燃烧的工作机理,甄选出蓄热室顶部温度(后简称蓄顶温度)作为焦炉火道温度软测量过程参数。然后,针对蓄顶温度和火道温度的复杂关系,建立线性回归模型来拟合两者问的线性关系;而对蓄顶温度和焦炉火道温度之间的非线性关系,本文将线性回归模型的预测误差,视为两者非线性关系的影响所致,采用基于时间差分法的Elman神经网络来拟合和多步预测,并采用规则集成的方法,将两模型组合成长周期组合预测模型。然而长周期组合预测模型的缺乏较好的实时性,本文继而在分析火道温度和蓄顶温度采样的基础上,采用多率采样的思想,对长周期组合预测模型有效和失效时,分别构建曲线拟合预测模型和多项式预测模型,使原焦炉燃烧系统由多率系统,成功转变成单率系统,从而全面改善集成模型的实时预测性能。最后,为了使软测量离线模型更能适应工况的变化,提出了一种模型自学习策略,进一步改进了集成模型的预测精度和应对焦炉燃烧工况突变的能力。建立在焦炉燃烧过程现场数据上的模型仿真,其对火道温度良好的拟合和预测效果,有力的佐证了基于多率采样的焦炉火道温度软测量集成模型的有效性和可行性。更多还原

【Abstract】 Coke is the main raw material in metallurgy industry, and is widely used in various industries. The combustion process is an important part in coking process and the flue temperature has direct impacts on the coke quality and resource consumption. However, it is difficult to measure the flue temperature through real-time method with low cost. And even the manual measurement is used, it also has the shortage of long circle and time-delay. The flue temperature has the characteristics of nonlinearity and time-varying, which is difficult to be measured. An integral model design of soft sensor based on multi-rate sampling is put forward to acquire the real-time flue temperature in this thesis.Based on the careful analysis of the background and significance of this subject, the thesis firstly decides to use the integral model, basing on the soft sensor, to analyze the coking combustion process. In the view of technical mechanism from coking process, the regenerator top temperature is selected, which to be used as the process parameter of the coke oven flue temperature’s soft sensor. Aiming at analyzing the complex relationship between regenerator top temperature and flue temperature, linear regress models (LR) are employed to reflect the linear relationship between them. Then in order to reflect the nonlinear relationship between them and predict multi-step ahead, the model combines the temporal difference method and Elman neural network (TD-ENN) is created. The TD-ENN is built on the basis of linear regress models’ errors and then, the LR and TD-ENN are integrated as the combined predictable model (CPM) with some important rules. However, when the CPM is created on the long period data, it lack of good real-time ability. Then the multi-rate method is used to solve the CPM’s deficiency. The curve fitting predictive model and the polynomial predictive model are built on the multi-rate method, consequently, when the CPM is valid or invalid, the shortage of the CPM is offset by one of the models respectively and the coking system is transforming from multi-rate to single rate successfully. At last, in order to fit the variable working condition well, the self-learning strategy is integrated in the combined model and the ability of the model is improved obviously. The effective drawings of the model simulation are built on the field data, the accurate precision of the prediction proves the high performance of the integrated coke oven flue temperature soft sensor model basing on the multi-rate.更多还原