Global Optimization of Process Models Incorporating Detailed Chemical Kinetics Mechanisms

摘要-CARR 9504827化学工程涉及复杂过程的设计、施工、操作和控制。相关的体系几何结构可能很复杂，因为所涉及的化学反应发生在时间和空间变化的环境中，和/或反应本身可能非常复杂，需要数百个物种参与更多反应的网络。系统或过程的优化可能需要通过高维参数空间进行全局搜索，搜索的每一步都需要将化学动力学方程以及质量和热传输方程与时间相关的边界条件和空间不均匀相结合。因此，对反应速率和计算能力的定量信息提出了很大的要求。这项研究将从两个方面着手解决这个问题：(1)开发数值分析和可视化工具，以及(2)将现有工具应用于不同复杂程度的动力学和反应堆模型。PI计划开发一种数据并行代码，用于在明尼苏达超级计算机研究所的CM-5(连接机)上对大的常微分方程组/微分代数方程(ODES/DAE)进行局部优化。已经开发了一种增广拉格朗日代码，并将编写另一种结合了连续二次规划的代码。这些程序将用于通过正交配置同时优化和求解底层的ODE/DAE，并用于研究更大和更详细的甲烷氧化动力学模型。一维输运和简单的湍流模型也可以包括在内。开发的下一阶段将是结合多开始随机搜索构建贝叶斯停止规则的代码。随后构建了甲烷小步氧化动力学机制。它将被优化，以匹配简单反应器中全动力学方案的输出，接近后者的全局最优值。

Abstract - Carr 9504827 Chemical engineering is concerned with the design, construction, operation, and control of complex processes. The associated system geometry can be complex because the chemical reactions involved take place within a time- and space-varying environment and/or the reactions themselves may be very complex, entailing hundreds of species participating in a network of many more reactions. Optimization of the system or process may require a global search through a parameter space of high dimension, each step of the search requiring integration of the chemical-kinetic equations and the equations of mass and heat transport with time-dependent boundary conditions and spatial inhomogeneity. Large demands are thus made for quantitative information on reactions rates and for computing power. This research will address the problem by proceeding along two fronts: (1) the development of numerical analytic and visualization tools, and (2) the application of the available tools to kinetic and reactor models of varying complexity. The PIs plan to develop a data parallel code for the local optimization of large sets of ordinary differential equations/differential algebraic equations (ODEs/DAEs) on the CM-5 (connection machine) at the Minnesota Supercomputer Institute. An augmented Lagrangian code has been developed and another code incorporating successive quadratic programming will be written. These codes will be tailored for simultaneous optimization and solution of the underlying ODEs/DAEs via orthogonal collocation and used to study larger and more detailed kinetic models for methane oxidation. One dimensional transport and simple turbulence models may also be included. The next stage of development will be to construct a code for Bayesian stopping rules in conjunction with multistart random search. The construction of a small step methane oxidation kinetic mechanism will follow. It will be optimized to match the output of the full kinetic scheme in a simple reactor in t he vicinity of the global optimum of the latter.