刮膜式分子蒸馏过程的优化控制研究与应用

发布时间：2018-03-11 19:45

本文选题：刮膜式分子蒸馏　切入点：预测模型　出处：《长春工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：分子蒸馏是一种新型的在高真空条件下进行的液-液分离技术,具有蒸馏温度低,真空度高,物料受热时间短,分离效率高等特点;且分离过程不可逆,没有沸腾鼓泡现象,特别适用于分离高沸点、热敏性、高粘度和易被氧化的物质。该技术已在医药行业、石油化工、食品工业、化妆品工业和农业等各行各业中得到了广泛应用。在分子蒸馏的现有的诸多研究当中,其建模、工艺参数的优化及控制量的控制算法研究较少,但是这些研究对于分子蒸馏的生产节能、提高生产的有效时间及减少对人工经验的依赖都起着很重要的作用。因此本文根据五味子粗油作为实验原料的提纯实验结果,对工艺参数与产品指标之间的关系进行了讨论,并结合刮膜式分子蒸馏的理论分析得到了影响被提纯物的纯度和得率的几个主要参数。然后在此基础上展开了对刮膜式分子蒸馏过程的建模、多参数优化以及控制量的控制算法的研究,本文的具体研究内容可以概括为如下几点:首先,提出了采用极限学习机建立刮膜式分子蒸馏过程的预测模型。由于分子蒸馏系统的非线性、强耦合、大滞后等特性,采用机理建模很难实现,并且考虑到BP网络的局部最优、训练时间长的缺点,因此提出采用极限学习机建立分子蒸馏系统的预测模型,仿真结果也表明该预测模型能够更加准确及时的预测系统的下一时刻输出状态。其次,采用启发式动态规划方法优化刮膜式分子蒸馏过程的多个参数,并对该算法进行了改进。由于参数间的耦合,单个参数优化效果并不理想,并且多个参数要进行多次优化,所以本文提出多参数同时优化方法。通过该方法能够在任意初始状态下,快速找到较好的工艺参数,缩短了参数调节时间。由于前人是采用BP网络实现启发式动态规划,将BP网络的如局部最优、调节时间长等问题也引入到了该算法中,因此本文给出了用极限学习机实现启发式动态规划的方法的具体理论推导,使改进后算法的寻优速度提高了近一倍。再次,提出了非线性系统的基于序贯极限学习机的逆模型控制方法。该方法不仅解决了解析逆系统难以求得的问题,还实现了逆系统的在线调整。该方法使用逆模型作为控制器,将逆系统与原系统的串联,逆系统的输出直接作用于被控对象。该方法能够有效的控制分子蒸馏系统的控制量如电机转速、热油机的温度控制等。最后,设计了刮膜式分子蒸馏的现场总线的控制方案,并且利用OPC(Object Linking and Embending,OPC)通信技术,能够将上位机中运行的高级算法的运算结果通过PROFIBUS网络发送到现场的控制器中。
[Abstract]:Molecular distillation is a new liquid-liquid separation technology under high vacuum conditions. It has the characteristics of low distillation temperature, high vacuum, short heating time and high separation efficiency, and the separation process is irreversible without boiling bubbling. It is especially suitable for the separation of substances with high boiling point, heat sensitivity, high viscosity and easy oxidation. The technology has been used in the pharmaceutical, petrochemical and food industries. It has been widely used in cosmetics industry and agriculture. Among the existing researches on molecular distillation, there are few researches on modeling, optimization of process parameters and control algorithm of control quantity. However, these studies play an important role in the production of molecular distillation energy saving, increasing the effective time of production and reducing the dependence on artificial experience. Therefore, based on the experimental results of crude oil from Schisandra chinensis as raw material, The relationship between process parameters and product indexes is discussed. Based on the theoretical analysis of scraping molecular distillation, several main parameters affecting the purity and yield of the purified product were obtained. Then, the modeling of the scraped membrane molecular distillation process was carried out. The research of multi-parameter optimization and control algorithm of control quantity can be summarized as follows: firstly, In this paper, a predictive model of scraping membrane molecular distillation is proposed by using the extreme learning machine. Because of the nonlinear, strong coupling and large delay characteristics of the molecular distillation system, it is difficult to use the mechanism modeling, and the local optimum of BP network is considered. Because of the disadvantage of long training time, a prediction model of molecular distillation system based on extreme learning machine is proposed. The simulation results also show that the prediction model can predict the output state of the system at the next moment more accurately and timely. Secondly, The heuristic dynamic programming method is used to optimize the parameters of the scraped membrane molecular distillation process, and the algorithm is improved. Because of the coupling between the parameters, the optimization effect of single parameter is not satisfactory, and many parameters should be optimized several times. Therefore, this paper proposes a multi-parameter simultaneous optimization method, which can quickly find better process parameters in any initial state, and shorten the adjusting time of parameters. Because the former use BP neural network to realize heuristic dynamic programming, The problems such as local optimum and long adjusting time of BP neural network are also introduced into the algorithm, so this paper presents the theoretical derivation of the heuristic dynamic programming method using the ultimate learning machine. The optimization speed of the improved algorithm is nearly doubled. Thirdly, the inverse model control method based on sequential limit learning machine for nonlinear systems is proposed. This method not only solves the problem that the analytical inverse system is difficult to obtain. In this method, the inverse model is used as the controller, and the inverse system and the original system are connected in series. The output of the inverse system acts directly on the controlled object. This method can effectively control the control quantities of the molecular distillation system such as the speed of the motor, the temperature control of the thermal oil engine, etc. Finally, the field bus control scheme of the scraper membrane molecular distillation is designed. And by using OPC(Object Linking and EmbendingOPC (OPC) communication technology, the result of the advanced algorithm running in the upper computer can be sent to the controller in the field through the PROFIBUS network.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TQ028.31

【参考文献】