Hybrid modelling and optimization of process systems

通过模拟预测过程系统的行为(制药、化工、炼油或类似的工厂)使它们能够进行优化设计和操作。这些系统由网络建模，其中节点(设备)转换能量或材料，而弧线是物质、能量或信息流。目前对过程系统建模的能力受到设备模型的准确性的限制，以及求解整个网络的算法。第一原理设备模型通常需要非常大的努力才能达到所需的精度。将第一原理和经验模型相结合的模型(所谓的混合模型)已经被探索作为替代方案。但目前还没有明确的方法来确定在混合模型的每个部分中包括什么类型的方程。目前，过程网络通过算法进行模拟或优化，这些算法本质上受限于单个CPU上的单个计算过程，并且没有利用多处理器的计算能力。研究的目标是通过以下方式使大型过程系统的模拟速度提高一个数量级，并显著减少建立准确的工厂设备模型所需的工作量：i.为特定类型的设备开发典型的混合模型结构-从而提供快速设备建模的手段。ii.过程网络的一种新的计算体系结构-通过自主行为并可以同时运行的节点模型进行模拟(基于代理的网络计算)。这项研究将为新一代炼油、聚合物、制药或生物应用化学品的工艺设计和工艺优化操作提供基础。

Prediction of a process system behaviour (pharmaceutical, chemical, refining or similar plants) via simulation has enabled their optimal design and operation.  These systems are modeled by networks where nodes (equipment) transform energy or material, while arcs are material, energy, or information flows.  Current abilities to model process systems are limited by the accuracy of equipment models, as well as the algorithms for solving the entire network.  First principles equipment models often require a very large effort to achieve desired accuracy.  Models combining the first principles and empirical models (so called hybrid models) have been explored as an alternative.   At present there are no well defined methods that determine what types of equations to include in each part of a hybrid model.   Currently, process networks are simulated or optimized by algorithms that are inherently limited by being single computational processes on single CPUs and do not exploit multi-processor computing capabilities.Goal of the research is to enable an order of magnitude improvement in simulation speed for large process systems and a significant reduction of the effort required to build accurate plant equipment models by:I. Development of typical hybrid model structures for specific types of equipments - thereby providing means for rapid equipment modeling.II. A new class of computational architecture for a process network - simulation via node models that behave autonomously and can run simultaneously (agent-based network computations),  andIII. Algorithms enabling optimization of discrete & continuous variables in such agent-based systems.  The research will provide a basis for a new generation of applications in process design and optimal operation of processes in refining, polymers, pharmaceuticals, or chemicals for bio applications.