【作者】 熊富强；

【导师】 桂卫华；

【作者基本信息】 中南大学 ， 控制科学与工程， 2012， 博士

【摘要】 1在常压富氧直接浸出过程中,硫化锌精矿含铁量达到5%～15%,浸出液中过量的铁元素会降低净化工段的除杂效率,甚至会导致后续生产的不稳定。针铁矿法沉铁通过在五个串联连续搅拌反应器中分别通入氧气和添加焙砂,使溶液中的铁离子以针铁矿的形式沉淀。它是一个氧化、水解及中和反应共同进行的多相化学反应过程,具有极强的非线性和耦合性。由于多相反应的复杂性和沉铁长流程的时滞性,使得过程控制难度大,低铁离子氧化速率和溶液pH值波动剧烈,导致生产过程稳定性差,能耗高,有价金属回收率低。因此,沉铁过程的建模和铁离子浓度预测,实现沉铁过程的优化控制,企业节能降耗优质高产的生产目标,具有重要意义。针对上述现象,本文在工业数据预处理和过程参数相关性分析的基础上,建立针铁矿法沉铁过程的机理模型、铁离子浓度的混合智能预测模型以及基于机理模型和智能模型的集成预测模型,并成功应用于针铁矿法沉铁过程的优化控制系统中。论文主要工作和创新性成果体现在以下几个方面：(1)针铁矿法沉铁过程的数据预处理以及相关性分析针对沉铁反应器出口pH检测值的过失误差所导致的建模精度低的问题,提出了基于多元回归的反应器出口pH值过失误差校正方法。针对串联反应器铁离子浓度缺失问题,提出了基于一阶动力学与平均插值的铁离子浓度插值方法。在针铁矿法沉铁机理研究的基础上,对操作变量,状态变量和生产指标进行相关性分析,为预测建模提供依据。实验结果表明,所提的数据预处理和分析方法能提高工业数据在建模过程中的适用性,改善了模型的精度。(2)建立了不同工况下的有效溶解氧浓度模型针对低铁离子氧化反应过程中有效溶解氧浓度欠缺所导致的机理建模困难的问题,提出了一种基于氧浓度差异的反应器分类方法。首先根据铁离子浓度在各沉铁反应器中的分布与变化,将沉铁反应器分成三类；然后根据氧气在溶液中溶解的影响因素,基于双膜理论渗透理论和表面更新理论,建立了不同工况下溶液中的有效溶解氧浓、度模型。该模型能客观准确地反映针铁矿法沉铁过程中沉铁反应器内的溶解氧浓度。(3)建立了沉铁过程的串联加权耦合机理模型针对针铁矿法沉铁过程反应机理的复杂性,首先基于化学反应动力学原理,分别建立了低铁离子的氧化反应、高铁离子的水解反应和氢离子的中和反应的动力学模型；然后根据质量守恒原理并结合动力学模型,建立了描述针铁矿法沉铁过程串联反应器的常规机理模型;在此基础上,提出了一种加权耦合机理模型,以补偿主反应间的耦合效应引起的常规机理模型计算误差。(4)提出了一种沉铁过程机理模型的参数辨识方法针对机理模型的强非线性,以及待辨识参数之间的耦合性,提出了一种基于分段进化PSO算法的模型参数辨识算法。同时,针对加权耦合机理模型中的权系数,提出了一种基于补偿特性的权系数确定方法,并分析了权系数随反应器的变化趋势。通过典型工况的数据仿真实验表明,该参数辨识方法能够有效地提高机理模型的精度。(5)生产指标铁离子浓度的集成预测模型针对机理模型预测精度低和沉铁后液铁离子浓度检测滞后的问题,建立了基于改进支持向量机和过程神经网络的信息熵集成预测模型,有效地解决了预测模型建模过程中样本点少和输入数据动态变化的问题。在实际生产过程中,针对该预测模型的精度在时变情况下的不稳定性,提出了基于改进欧式距离选择样本的OLS局部建模方法,以及基于信息熵集成预测模型性能的动态加权集成预测模型。另外,从提高沉铁后液铁离子浓度预测模型在不工况下预测精度的角度出发,建立了一种基于沉铁过程多工序多元过程能力指数的机理模型和智能模型的铁离子浓度集成预测模型,最后通过现场数据的仿真结果验证了该集成预测模型的有效性。(6)开发了针铁矿法沉铁过程的优化控制系统该系统通过OPC客户端实现现场数据采集,在此基础上,实现了针铁矿法沉铁过程的关键工艺参数的协调优化设定、沉铁后液离子浓度的在线预估以及操作变量的优化调节等功能。另外,还实现了针铁矿法沉铁过程的流程监视、数据查询和分析等功能。实际工业运行结果表明该系统有效地提高了企业的生产效率和资源利用率。更多还原

【Abstract】 In the directly leaching process under atmospheric pressure with rich oxygen, zinc sulfide concentrate often contains significant iron irons (5%～15%), which decreases efficiency of impurity removing in the next stage (purification), and could even lead a unstable production station. In iron precipitation process, iron ions are deposited as goethite in five continuous stirred tank reactors (CSTR) in series using zinc calcine and oxygen. In this process, multi-phase chemical reactions, including oxidation, hydrolysis and neutralization, display at the same time with serious coupling and nonlinearity. For the complexity of multi-phase reactions in the process, it is not easy to control iron precipitation process, and the oxidation rate of ferrous ion and the pH value fluctuate seriously, leading low stability of production process, high energy consuming and low valuable metal recovery. Hence, the modeling of iron precipitation process and prediction of iron concentration are significant to the optimization control of the process, and benefit to energy saving, high production of the company.To solve the above problems, the optimization calculation in iron precipitation by goethite process, series weighted coupling CSTR model, hybrid prediction model for iron ion concentration and integrated prediction model for iron content have been built, in the basis of pre-processing and relevance analysis of industrial data. The major work and the innovative achievements in this paper are reflected in the following areas:(1) Data preprocessing and relevance analysis of iron precipitation by goethite processTo solve the problem of low modeling accuracy caused by gross error in detected value of pH meter, the gross error correction method of outlet pH value based on a multiple regression and first order kinetic was presented. And aimed at the undetection of iron concentration of the central reactor in CSTR, the interpolation method for iron concentration combined first order kinetics with average interpolation algorithm was proposed. On the basis of mechanisms study of iron precipitation as goethite, the relevance among the operating variables, state variables and production targets were analyzed. The experimental results showed that the proposed data pre-processing and analysis method promoted the efficiency of industrial data in modeling and increased the accuracy of the models(2) Effective dissolved oxygen concentration model under different working conditionsIn the view of the modeling problems caused by lack of effective dissolved oxygen concentration, the iron precipitation reactors classification method according to oxygen dissolving differences was proposed. In the method, iron precipitation reactors were classified into three groups after analysis of iron ion concentration distribution and changes in each reactor. Then according to the influences of oxygen dissolving in the solution, the effective dissolved oxygen concentration model under different working conditions on the basis of the two-film theory, penetration theory and surface renewal theory, has been analyzed and built. The proposed model could describe the dissolved oxygen concentrations objectively in each iron precipitation reactors.(3) Series Weighted coupled CSTR modeling in iron precipitation by goethite processFor the complexity of reaction mechanism in iron precipitation as goethite, kinetic models of oxidation, hydrolysis and neutralization were established, respectively, according to the principles of chemical dynamics. Combined with the kinetic models, the regular series of continuous stirred tank reactor model (n-CSTR) was built based on mass conservation. On the basis of those work, the weighted coupling CSTR model was proposed to compensate calculation errors of regular mechanism model caused by the coupling effect among the three reactions.(4) An iron precipitation mechanism model parameters identifica-tion method is put forwardAccording to the strong nonlinear of mechanism model and the coupling between the identified parameters, this paper proposes a model parameter identification algorithm based on the piecewise evolution PSO algorithm. Aiming at the weight coefficient of weighted coupling mechanism model, a method to determine the weight coefficient has been proposed in this paper, and the change tendency of the weight coefficient along with the reactor has been analyzed. The data simulation experiment of typical working conditions show that the parameter identification method can effectively improve the accuracy of the mechanism model.(5) Integrated prediction model for iron ion concentration in production targetFor the mechanism model prediction accuracy and real-time difference, as well as the time lag of low iron and ferric ion detection, an intelligent forecasting model based on improved support vector machine and process neural network ensemble has been built to efficiently solve the problem that the sample point was too little and input data dynamically changed during the iron precipitation by goethite.According to the prediction accuracy of the forecast model in the production process time-varying situation, this paper has proposed a Euclidean distance sample selection method based on the relevance analysis and weighing, in addition, a correction method for prediction model based on the capability of prediction model and local regression.In addition, aiming at raising the precision of prediction model under different work conditions, a integrated model including mechanism and intelligent model is presented based on the mulstitgae process multivariate quality process capability index size. The scene data simulation results verified the validity of the integrated prediction model.(6) Developing the optimization control system of iron precipita-tion by goethite processThrough the OPC client this system can collect scene data and realize the function of the key process parameters coordination optimization setting in the iron precipitation by goethite process, ion concentration online estimate after precipitation process and operating variable optimization control. In addition, it also can achieving process monitoring, data query and analysis function in the iron precipitation process. The actual industrial operation results confirmed that this system efficiently improved the enterprise’s production efficiency and the resources utilization.更多还原