基于流股和单元的热力学物性建模和计算方法研究

发布时间：2018-03-02 01:30

本文关键词： 物性计算代理模型流股和单元自动化实现空气分离过程乙烯淤浆聚合　出处：《浙江大学》2015年硕士论文　论文类型：学位论文

【摘要】：热力学物性计算是化工过程模拟与优化计算的基础,它对于流程的收敛性能和求解结果有着至关重要的影响。依据化工对象建立的机理模型通常都会涉及到流股或是操作单元的物性数据,这些物性参数一般包括了相平衡常数、密度、焓、熵等。在对模型进行模拟计算前,首先要确定精确可靠的物性计算方法。但在实际体系中的物性计算往往比较复杂,可能存在以下难点：①使用联立方程法建模时,某些涉及到条件判断、序贯迭代等过程的物性计算过程破坏了模型的全联立；②使用代理模型模拟时,求解器无法获得迭代点的梯度信息；③大规模过程模型中需要建立数量众多的物性计算模块,建模的成本巨大。针对以上问题,本文的主要研究工作和贡献如下：1、基于Kriging回归方法和LHS采样方法,提出了对热力学物性模型建立代理模型的方法,研究了物性模型输入输出特征并结合Kriging回归方法选择合适的回归模型和参数,使得针对不同的物性体系都能得到精度较好的代理模型。2、为了使得物性代理模型有较好的收敛性,提出了代理模型的梯度模型,通过改写回归模型的解析导数得到代理模型的梯度信息,并传递给建模平台下的求解器,使得加入了物性代理模型的全联立模型依然保持很好的收敛性。3、针对化工对象流股和操作单元众多的情况,提出了基于流股和单元的物性建模方法,可以根据对象物性体系中工况和组成情况快速建立物性代理模型,且满足最大化复用性和相互独立性,降低了建模的成本。4、为了提高建立物性模型的效率,开发了一套自动化程序和外部函数模型,用户可以根据需求完成物性代理模型的参数回归工作并自动生成所需的参数文件,只需将外部函数语句写入原模型中并关联参数文件即可实现物性代理模型的计算。5、针对空气分离过程和乙烯淤浆聚合过程,使用了物性代理模型构建模型中的物性计算部分,克服了原有物性计算方法存在的条件判断、序贯迭代等问题,保证了模型的全联立,通过模拟结果验证了物性计算较好的准确性和收敛性。
[Abstract]:The calculation of thermodynamic properties is the foundation of chemical process simulation and optimization calculation. It plays an important role in the convergence performance of the process and the solution results. The mechanism model based on chemical objects usually involves the physical property data of the flow stock or the operation unit, and these physical parameters generally include the phase equilibrium constant. Density, enthalpy, entropy, etc. Before simulating the model, it is necessary to determine the accurate and reliable calculation method of physical properties. However, the calculation of physical properties in practical systems is often complicated, and there may be the following difficulties: 1, when modeling by simultaneous equation method, Some physical properties calculation processes involving conditional judgment, sequential iteration and so on destroy the whole simultaneous modeling of the model when the agent model is used to simulate the model. The solver can't get the gradient information of the iteration point in the large-scale process model, which needs to build a large number of physical properties calculation modules, and the cost of modeling is huge. The main research work and contributions of this paper are as follows: 1. Based on Kriging regression and LHS sampling method, a method of establishing agent model for thermodynamic properties model is proposed. This paper studies the input and output characteristics of physical property model and selects appropriate regression model and parameters combined with Kriging regression method. The agent model .2with good precision can be obtained for different physical property systems, in order to make the physical agent model have better convergence. The gradient model of the agent model is proposed. The gradient information of the agent model is obtained by rewriting the analytic derivative of the regression model, and the gradient information is transferred to the solver under the modeling platform. The whole model with physical agent model is still convergent. 3. In view of the large number of flow stocks and operation units of chemical objects, a physical property modeling method based on flow stocks and units is proposed. The physical agent model can be established quickly according to the working condition and composition of the object physical property system, and it can satisfy the maximization of reusability and mutual independence, and reduce the cost of modeling .4. in order to improve the efficiency of building the physical property model, A set of automatic program and external function model are developed. The user can complete the parameter regression work of the physical agent model according to the requirement and automatically generate the required parameter file. The calculation of physical property agent model can be realized by writing the external function statement into the original model and associating the parameter file. For the air separation process and ethylene slurry polymerization process, the physical properties calculation part of the physical property agent model is constructed by using the physical property agent model. It overcomes the problems of conditional judgment and sequential iteration in the original physical property calculation method, and ensures the whole simultaneous operation of the model. The simulation results verify the accuracy and convergence of the physical property calculation.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ013.1;TQ019

【参考文献】