【作者】 朱红求；

【导师】 阳春华；

【作者基本信息】 中南大学 ， 控制科学与工程， 2010， 博士

【摘要】 湿法炼锌工艺是世界上锌冶炼的主要工艺,采用该工艺生产的锌占世界总产量的80%以上。净化除钴过程是湿法炼锌工艺的一个重要工序,主要通过添加锌粉置换除去硫酸锌溶液中的钴离子杂质。由于净化除钴过程机理复杂,流程长且影响因素多,特别是杂质离子浓度无法在线检测,导致过程优化控制困难,净化除钴过程锌粉消耗大。论文在分析净化除钴过程影响因素和过程信息特点的基础上,研究了基于数据的过程建模与优化控制方法,构建了过程优化控制总体结构。在过程异常数据检测、缺失数据补全的基础上,提出了基于时间序列分析的入口溶液离子浓度在线估计方法,建立了基于支持向量机的净化除钴过程工艺指标预测模型,在此基础上研究结合案例推理预设定和支持向量机补偿修正的优化控制技术,并将其成功应用于净化除钴过程的控制系统中。论文主要研究工作及创新性成果如下：(1)提出了净化过程数据预处理技术。在分析净化除钴过程异常数据产生原因的基础上,提出了基于局部平均距离和工况参数预估的数据异常检测方法。针对过程数据丢失、异常等原因造成数据缺失问题,研究了基于案例推理的缺失数据补全方法。采用工业现场过程数据进行模拟分析与验证,结果表明所提方法有效提高了过程数据质量,为基于数据的净化除钴过程建模与优化创造了条件。(2)提出了基于时间序列的入口溶液离子浓度在线估计方法。利用小波分解的方法,将离子浓度人工分析检测时间序列分解为多个子序列,通过相空间重构,在重构的子空间建立支持向量机模型。将子序列模型输出重构,实现了净化除钴入口溶液离子浓度的在线估计。其中支持向量机的参数采用混沌粒子群算法进行优化。应用工业生产数据验证的结果表明,相对误差小于10%的样本达97.5%,模型在线估计精度满足现场实际生产工艺要求,为净化除钴过程的优化控制提供了可靠信息。(3)建立了基于模糊聚类和支持向量机的工艺指标预测模型。针对净化除钻过程数据量大、过程具有很强的非线性等特点,研究了基于加权模糊聚类的分类方法,将样本空间进行分类；在分类得到的每个类中,建立了各个类的模糊支持向量机子模型,综合考虑样本数据时间域和空间域的影响,设计了一种复合模糊隶属度函数；在此基础上,将子模型输出集成,得到了过程指标预测模型的输出。其中,采用递阶粒子群算法对于属性选择和权重进行优化。研究了模型的校正方法,提高了模型的精度。现场数据实验验证了该方法的有效性。(4)提出了基于案例推理的预设定模型以及基于支持向量机的补偿模型,实现了净化除钴过程的优化控制。该方法依据在线估计模型提供的离子浓度信息,结合入口溶液流量,利用案例推理策略建立了锌粉添加量的预设定模型,有效克服了浓度、流量等因素对优化控制的影响。在此基础上,根据过程指标预测模型计算的出口离子浓度预测值与期望目标之间的偏差,建立了基于支持向量机的锌粉添加量补偿模型。实验结果表明,所提出的优化控制策略有效减少了锌粉的添加量,节约锌粉6.48%。(5)开发了净化除钴过程的优化控制系统。该系统利用OPC技术实现与现场DCS系统之间的数据通信,实现了净化除钴过程的入口溶液离子的浓度在线估计、过程工艺指标预测以及锌粉添加量操作优化等功能。此外,还实现了净化过程全流程监视、数据查询以及过程报表分析等功能。实际工业运行结果表明了该系统的有效性和可行性。更多还原

【Abstract】 Above 80% of total yield of zinc is prodecued by hydrometallurgy in the world. The cobalt removal in purification process is the critical step in zinc hydrometallurgy, where the cobalt in the zinc sulfate solutions is removed by cementation reaction using zinc powder. The cobalt removal mechanism is complex and there are many factors that influence the removal effects. Especially, the ion concentration of the zinc sulphate solution cannot be measured online. These factors lead to the difficulty in operation parameters optimization and more consumption of Zinc powder in cobalt removal processs.Based on the analysis of influencing factors and the characteristics of the process data in the cobalt removal process, the process modeling and optimization control strategy is studied with industrial data and the whole structure of optimization control technology is proposed. Firstly, a preprocess method of process data including the outlier identification and imputation of missing data is proposed. Then, an on-line estimation method of the influent ionic concentration is researched based on time series analysis. Finally, the process index prediction model based on SVM is estabilished and an optimization control strategy for cobalt removal process combining the preset model based on case-base reasonning and SVM-based compensating model is presented. In addition, the optimization control system is developed based on the proposed methods mentioned above. The major innovation research achievements include:(1) The data preprocess technique of process data was proposed. On the basis of analysis of the main source of abnormal data, the detection method of abnormal data is studied by using the local average distance and estimation of technological situation parameter. Aimed at the problem of the missing data caused by lossing data and abnormal data, the imputation method was proposed based on case-based reasoning. The verified results of on-site industrial production data showed that the proposed data preprocess method can effectively improve the process data quality and provide the condiction for modeling and optimization control of cobalt removal process based on data-driven.(2) An on-line estimation method of the influent iconic concentration was proposed for the cobalt removal process based on the time series analysis. Using wavelet decomposition, the time series of ionic concentration were decomposed into multiple subsequences and these subsequences were reconstructed into each subspace in phase space. Then, the support vector machine (SVM) model was built in each subspace. Finally, the outputs of each subsequence model were synthesized as the result of on-line estimation model for the influent iconic concentration in the cobalt removal process. The parameters in SVM models were optimized by using a chaotic particle swarm (PSO) algorithm. The verified results of the industrial production data showed that the samples with relative error less than 10% in the proposed model is up to 97.5%. It also indicated that the on-line estimation precision met the technological requirement of practical industrial production. The proposed model provides reliable information for the optimization control of cobalt removal process.(3) A prediction model of technology index for cobalt removal process is built based on fuzzy c-means (FCM) clustering and fuzzy SVM. According to characteristics of the non-linearity and large-scale production data of cobalt removal process, the weight FCM clustering algorithm was used to separate the whole training data set into several clusters. Then, each cluster subset was trained by fuzzy SVM respectively and obtains its corresponding sub-model. These fuzzy sub-models use a hybrid fuzzy membership, in accordance with the different effect of the sample in different periods and in different sample space. Finally, a weighted integrated model was formed with these outputs of sub-models. A hierarchical PSO method is used in feature attributes selection and weight vector optimization. In addition, a correction method is used to improve the model precision. The experimental results of industrial data demonstrated the effectiveness of the proposed prediction method.(4) An optimization control strategy of zinc powder addition amount is studied, which integrates the optimal preset model based on case-based reasonning with the compensation model based on support vector machine. Using the ionic concentration information provided by on-line estimation model and the influent solution flow rate as input variables, the optimal preset model based on case-based reasonning determines the preset value of zinc powder addition amount, which effectively overcomes the influences of the fluctuations of ionic concentration and the influent solution flow rate. Then, according to the deviation between the expected set-point value and the feedback prediction value of the effluent ionic concentration, the compensation model of zinc powder addition amount is proposed based on SVM. The experimental validation results showed that the proposed optimization control strategy can effectively reduce the zinc powder addition amount by 6.48% in average.(5) An optimization control system for the cobalt removal process has been developed. The data communication between the developed system and the distributed control system (DCS) is realized with OPC technology. It realized the functions including the on-line estimation of the influent iconic concentration, process technological index prediction, operation optimization of the zinc powder adding, as well as the functions of process monitoring, data inquiry and process report analysis and so on. The practical running results showed the effectiveness and feasibility of the developed system.更多还原