Transforming Equation-based Models in Process Engineering.

Abstract

In the solution of realistic dynamic optimization problems, the computation of derivative information is typically among the crucial ingredients in terms of both numerical accuracy and execution time. The aim of this work is to bring together automatic differentiation with the DyOS framework for dynamic optimization used in process engineering. In this framework, the optimization algorithms and the mathematical models describing the engineering systems are separated. In a realistic scenario, an engineering system is composed by integrating different submodels, possibly formulated using different modeling languages such as Aspen Plus, gPROMS, or Modelica. The DyOS is currently redesigned to be capable of handling such component-based models by relying on a common intermediate format called CapeML. The idea of CapeML is to define a layer of abstraction, so that various models can be expressed in a manner independent from a specific modeling language. So, CapeML is the adequate format to which automatic differentiation is applied in this dynamic optimization framework. A novel system called ADiCape is proposed, implementing the forward mode for models written in the XML-based language CapeML. This AD transformation is expressed in form of an XSLT stylesheet.