|
|
|
|
|
|
|
Abstract
We consider how Simulation can be used to design a system to yield optimal expected behaviour. We assume that the specific measure of the system behaviour ( goal function) depends on values of the parameters ( input parameters) chosen for the system. We want to determine the values of these parameters yielding the optimal value of this measure, possibly subject to some constraints. The underlying system is so complex, that there is no mathematical closed expression for the goal function and it is necessary to use simulation to evaluate it for every set of input parameter values.
The technique used for optimisation is expected to be able to locate the global optimum, to be robust in converging, not sensitive to noise in evaluations of single values of the goal function and efficient.
In the tutorial we present an introduction to simulation optimisation techniques.
We briefly review gradient techniques and focus on random search techniques.
The aim is to provide an introduction to major developments in the field of simulation optimisation, rather than a survey of the current status of optimisation or providing description of algorithms ready to work.
First we present a method for using simulation to estimate a gradient of the expected goal function with respect to input parameters values. Then we discuss stochastic approximation and sample path optimisation.
Finally we present two techniques base don random search: Simulated Annealing and Genetic Algorithms. We conclude with an introduction of a recent development, an optimisation tool REMO, which illustrates the use of advanced and efficient version of Genetic Algorithms for simulation optimisation.
Short vita
Dr. Szczerbicka has been graduated in Applied Mathematics and has got a Ph.D. in Computer Science from the Warsaw University of Technology, Poland. After some years of teaching and research at the same University, she was appointed at the University of Karlsruhe, Germany, then took a professor position at the University of Bremen, Germany. Since 2000 she has been chairing the Institute of the Computer Science at the University of Hannover, Germany as a full professor. She works actively in the Society for Computer Simulation (SCS), the Federation of European Simulation Societies (Eurosim) and the German Simulation Society (ASIM). She serves as a member of an editorial board of the book series Frontiers of Simulation. Recently she has been involved in international activities on establishing a curriculum for an educational program for a simulation degree and a certification of simulation professionals as a member of the international Modeling & Simulation Professional Certification Implementation Group, Washington, USA. Since 1990 she is an Associate Member of the McLeods Institutes of Simulation Sciences, USA. Since 1992 she has been conducting the FG4.5.3 Working Group Simulation and Artificial Intelligence of the German Society of Computer Science (GI).
She has been involved in organization of numerous conferences in the field of simulation, working in program committees, and as a track, program or general chair as well. She is an author or co-author of more than 80 papers and co-author of one book on Simulation and Artificial Intelligence. Four papers have got a label of a best paper award. Her research interests primarily focuse on the field of modelling performance and performability of large scale, discrete dynamic systems. In further development she expanded areas of interests onto application of methods of artificial intelligence to modelling, simulation and validation as well as distributed and web-based simulation.
Directions in her research have been frequently triggered by industrial co-operations.
Currently she has been involved in establishing a research Center for Integrated Simulation CIS at the University of Hannover, Germany.
Selected Publications
[30] Szczerbicka,H.;Syrjakow, M.: