【作者】 夏陆岳；

【导师】 潘海天；

【作者基本信息】 浙江工业大学 ， 化学工程， 2004， 硕士

【摘要】 数据挖掘技术是当今智能系统理论和技术的重要研究内容，它综合运用人工智能、计算智能、模式识别和数理统计等先进技术，从数据库中提取隐含在其中的有用信息和知识。本文充分利用数据挖掘技术的优点，对氯乙烯悬浮聚合过程进行了广泛研究，在PVC质量建模、聚合过程动态建模和引发体系优化等方面取得了许多研究成果，并在此基础上，提出了基于技术集成的PVC智能制造技术。 首先对数据挖掘技术在化工过程中的应用现状进行了综述，介绍了一些主要的数据挖掘研究方法，其中包括神经网络、遗传算法、遗传规划、主元分析、偏最小二乘法和粗集理论等，展望了数据挖掘技术在化工过程的应用前景，以及未来发展所需解决的问题。 针对颗粒特性这一PVC树脂的重要质量指标，采用多元回归分析方法建立了PVC平均粒径的回归模型，并运用趋势面分析方法研究不同搅拌转速下，纳米CaCO₃微乳液浓度和分散剂浓度等优化操作条件。 提出了基于组合神经网络—岭回归方法（SNNs-RR）的PVC颗粒特性预测方法，采用岭回归方法来选择合适的组合权重。通过与单一神经网络模型比较，结果表明采用SNNs-RR方法建立的组合模型具有更佳的预测精度和鲁棒性。 通过对反应器内反应混合物和冷却回路的热传递平衡进行分析，建立了PVC反应器的动态热量传递模型。由于考虑了反应过程中反应物质量及其比热的变化，所建的模型能准确地描述氯乙烯悬浮聚合过程的动态特性。运用此模型对氯乙烯悬浮聚合过程的反应速率、转化率和注水速率等过程状态和操作条件进行了必要的分析。 为了充分利用PVC反应器的冷却能力，氯乙烯悬浮聚合过程的反应速率最好能维持在一个与聚合反应系统最大冷却能力相符的水平上，解决该问题的常规方法是采用高低效引发剂相结合的复合引发体系。通过分析氯乙烯悬浮聚合过程的反应动力学，提出了引发体系的优化模型，并成功应用于二元复合引发体系的优化。 根据PVC智能制造技术的目标是将氯乙烯悬浮聚合过程的动态模型、新型仪表、多元统计方法及先进控制算法等反应控制手段实现于协调的系统集成中，提 浙江工业大学硕士学位论文出了基于技术集成的PVC智能制造技术，其中涉及到氯乙烯悬浮聚合过程模拟技术、先进控制与软测量技术、过程性能监测与故障诊断技术等。 本文充分运用信息技术为优化聚合配方和工艺条件，为提高PVC产品质量提供了有效的研究手段，有助于进一步开展聚合过程先进控制、在线优化和故障诊断等方面的研究，有利于促进智能制造技术在其它聚合物生产中得到进一步应用，为数据挖掘技术在化工过程中应用提供一种示范。更多还原

【Abstract】 Data mining, a rising technique nowadays, combines many advanced techniques, such as artificial intelligence, computational intelligence, pattern recognition, statistic etc., to mine and discover valuable and hidden knowledge from databases. Vinyl chloride suspension polymerization is widely studied by making full use of data mining in this paper. Many achievements about quality modeling of polyvinyl chloride (PVC), dynamic modeling of polymerization and optimization of initiation system are gained, and intelligent manufacturing technology (IMT) of PVC based on technology integration is presented.Firstly, the status of research and major achievements in the research of data mining technique in the chemical process are reviewed. Some primary methods in data mining are introduced including methods of artificial neural network (ANN), genetic algorithm (GA), genetic programming (GP), principal component analysis (PCA), partial least square (PLS), rough set (RS), etc. The application prospect, existing problems which needed solving for further developing are analyzed.Aimed at the importance of PVC particle feature, the regression model of the mean grain diameter is developed by using multiple nonlinear regression. The optimum operational conditions about the concentration of dispersant and nano-CaCO3 microemulsion is attained by using trend surface analysis.Inferential estimation of PVC quality using stacked neural networks - ridge regression (SNNs-RR) approach is studied, and determination of appropriate weights for combining individual networks using ridge regression is proposed. Neural network generalization capacity can be improved by combining several neural networks. The results obtained demonstrate significant improvements in model accuracy and robustness, as a result of using SNNs model, compared to using single neural network model.Based on heat transfer process form the reactants to the reactor wall and from the inner wall to the coolant in the jacket, the dynamic heat balances for the PVC reactor is developed. Since the changes in both the mass of the reactants in the reactor and its specific heat are considered, the model can be used to exactly describe the dynamic characteristic of vinyl chloride suspension polymerization. The analysis of the reactionrate, conversion and the rate of water draining is provided.In order to take full advantage of the PVC reactor cooling capacity, the polymerization rate must be kept approximately constant at a level which yields a reaction rate equal to the common maximum cooling capacity of the system. The common solution to this problem is to use a combination of rapid and slow initiators. The optimal formulation of initiation system is studied as an optimization problem based on reaction kinetics. The optimization model of initiation system is developed, and optimization of initiation system containing two initiators is successfully implemented.IMT of PVC based on technology integration is presented. Its objective is the harmonious integration of the dynamic model of vinyl chloride suspension polymerization, advanced sensors, multivariable statistical approaches and advanced process control strategy. It relates to dynamic simulation technique of vinyl chloride suspension polymerization, advanced control and soft sensor technique, process performance monitoring and diagnosis, etc.An effective research approach about improving quality of PVC by using information technology is provided. The results might be helpful to correlative study (advanced control, on-line optimization, diagnosis, etc.) in vinyl chloride suspension polymerization. The achievements also provide a basis for the further application of IMT in the production of others polymer, and provide an instantiation for the application of data mining in chemical process.更多还原