Simulation and Optimization as Modular Tasks within a Framework on High-Performance-Computing-Platforms

In the context of the project „Effizienter Flughafen 2030“ separate, autarkic models simulating the different processes at an airport have been coupled to build an integrative over-all model. For scenario analysis and to identify the relevant model parameters for the over-all behavior it was postulated to fit input parameter values automatically with regard to a certain optimal behavior of the coupled system. To solve this task, a framework approach had been developed using softcomputing optimization methods. Independently of its original motivation and application, this approach can be applied to all complex optimization problems, if an encapsulated specification of the model components and the optimization is provided. The soft computing approach has the advantage to be able to deal easily with typical phenomena known in the area of environmental modeling such as lack of definition, non-linearity, or incomplete data. On the technical level it was a quite natural step to transfer the concept of the model coupling to the complete framework including the optimization component and to implement the complete functionality as a web service. For the prototype introduced in this paper genetic algorithms are selected for the optimization module. The architecture of the genetic solver demonstrates the general framework approach and is able to use high-performance-computing platforms to produce considerable speedups without demanding high bandwidths or low latency interconnections for the platform used. A simple application example will demonstrate the interfaces for the model and the genetic solver that have to be provided to encapsulate these components for usage in the framework. The outlook deals with the optimization of the parameters of the genetic solver itself and proposes to interpret the black-box web-service interface of the solver as a model itself such cascading two genetic solvers: the inner one to optimize the simulation model parameters and the outer one to optimize the genetic algorithm parameters of the inner one.