【作者】 商秀芹；

【导师】 孙优贤； 卢建刚；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2010， 博士

【摘要】 烧结生产在我国钢铁企业中有着重要的地位。它是高炉进料的一个重要处理过程,能够最大程度地减小高炉原料的波动,为高炉的平稳生产提供可靠保障。同时,由于烧结矿是高炉的主要进料,因而烧结矿的质量直接关系着高炉生产的产量和质量。此外,随着铁矿石价格的攀升和全社会节能环保意识的增强,在钢铁工业中展开面向节能降耗的技术改进势在必行。作为影响高炉产量和质量的重要环节,烧结过程技术改进在整个炼铁过程节能降耗技术改进中必不可少。因此,烧结过程的研究具有重要理论价值和实际意义。本文研究了烧结过程的两个主要部分烧结矿配料系统和烧结过程热状态过程的建模与优化问题,采用改进的遗传规划算法(Genetic Programming, GP)建立问题的解析模型,通过非支配排序遗传算法(Non-dominated Sorting Genetic AlgorithmⅡ, NSGAII)的改进算法解决优化问题。本文的主要工作和贡献如下：(1)提出了分层辨识遗传规划算法。该算法针对被辨识系统各部分可辨识度不同的问题,采用分层辨识和反馈校正的基本思想,对研究对象分层辨识,直到辨识误差缩小到可接受的范围。同时,采用最小二乘法辨识模型参数,应用M估计技术增加模型抗干扰能力。通过实验仿真验证了该算法所建模型的精度优于一般遗传规划。(2)提出了混合分类遗传规划算法。混合分类遗传规划算法针对复杂的多工况系统的模型辨识问题,采用分类算法对系统输入输出数据进行分类,然后对每一类数据分别建立模型,从而建立起能描述每个工况的整体模型。同时,采用最小二乘法辨识模型参数,应用M估计技术增加模型抗干扰能力。通过实验仿真验证了该算法所建模型的鲁棒性优于一般遗传规划算法。(3)提出了一种带偏好的非支配排序遗传算法。该算法通过设置目标函数的期望值,在经典的非支配排序遗传算法中引入偏好信息。偏好信息引导搜索方向趋于决策者偏好的区域,增强算法在偏好区域的搜索能力。实验仿真分析了在多目标优化问题中带偏好的非支配排序遗传算法的优势。(4)建立了烧结矿化学成分和质量预测模型,并基于预测模型提出多目标配料优化模型。预测模型采用了分层辨识遗传规划算法进行离线建模,同时把基于部分机理的模型结构加入到初始种群中,使得部分机理与数据驱动的建模方式相互结合。采用现场数据对预测模型进行测试,并从理论上分析了模型误差的来源。在烧结过程配料多目标优化模型中,优化目标为全铁、碱度和转鼓强度波动最小。为求解该优化问题,采用基于偏好的非支配排序算法,偏好信息的引入增强了算法搜索能力,有利于决策者作出决策。利用现场数据验证了该优化配料系统的有效性。(5)为了控制烧结过程热状态,建立了烧结机风箱温度的两级预测建模,并通过该模型预测烧结终点。中期预测是在风箱温度拐点处建立风箱温度的自回归模型；短期预测是提前风箱温度最高点约两个风箱处建立风箱温度的自回归模型。在中期预测模型和短期预测模型中,通过二次和三次曲线拟合温度点并计算出曲线最高点即为烧结终点。利用现场数据对模型进行测试并分析了误差产生原因。更多还原

【Abstract】 Sintering is important for the iron-making industry. As an essential pre-process for blast furnace materials, it could minimize the fluctuations of the materials, and thus guarantee the furnace’s steady operation. With raising price in iron ore and public enhancing consciousness of energy saving and environmental protection, it is imperative to improve the technology in iron making plants in terms of saving energy and reducing cost. In this technical innovation, the sintering process has attracted more and more attention as it plays an important role. Therefore, it is practically significant to study the iron ore sintering process.Automatic control system for sintering contains mainly two parts:control of sinter mix proportion and control of thermal state. Both of them should be designed on the basis of the model of sintering process. However, it is difficult to establish the accurate mathematical model of sintering process. In this thesis, improved Genetic Programming algorithms are proposeed to conquer the above difficulty Then, in order to solve the multi-objective optimization problem in sintering process, Non-dominated Sorting Genetic Algorithm II (NSGAII) is applied and then enhenced.In detail, the major contributions of this thesis are summarized as followings:1. An improved Hierarchical Genetic Programming (HGP) is proposed to solve the problem of modeling the nonlinear dynamic system with unknown mechanism and massive data in industrial processes. Based on the hierarchy model and feedback, the algorithm identifies the model according to the hierarchies until the error can be acceptable. Least square method (LSM) and M-estimation is adopted to overcome the large noise and increase the robustness of the model. The experimental results demonstrated that this algorithm is effective in nonlinear dynamic model identification.2. An improved Classified Genetic Programming (CGP) is proposed. K-Means clustering or K-Medoid clustering is applied to partition conditions of the target objects. Besides, for each clustering, GP is proposed to construct the empirical model. CGP adopts least square method (LSM) and M-estimator to improve the abilities of computing and disturbance resistance. Simulation proved that CGP has satisfactory performances.3. A Preference-based Non-dominated Sorting Genetic Algorithm (PNSGA), an improved method of NSGA II (non-dominated sorting genetic algorithm II), is proposed to solve the multi-objective optimization problems. A new preferable relationship was defined based on Pareto dominance and combined with the fast non-dominated sorting. The advantage of our algorithm over NSGA II in terms of crowding mechanism was analyzed. Simulation results demonstrated the effectiveness of the algorithm on parameters identification of dynamic model, compared with conventional experienced methods.4. To avoid fluctuations of components and quality of iron ore sintering, quality prediction models are created by HGP, which contain the components and tumnler models of sinter. Partial mechanism knowledge is introduced into initial population of HGP, and the models are based on both the partial mechanism and the data. Simulation results showed the superiority of these quality predictive models. One optimization scheme of sinter mix proportions is proposed based on predictive quality of iron ore sintering. The optimization objectives are to minimize the fluctuation of TFe component, basicity and tumnler. PNSGA is utilized to solve the multi-objective optimization problems. Simulation results showed the superiority of sinter mix proportion based on quality predictive models.5. For predicting burning through point (BTP), two models for temperature prediction are established by CGP in the sintering process. The two models were the medium-term model based on the temperature inflexion and the short-term model based on the temperature neighboring to BTP. BTP was obtained by the cubic curve fitting of the predicted temperature. Simulation results proved the superiority of the two-term prediction model.更多还原