Research: Skövde researcher aims at finding the best solutions through Multi-objective Optimization
2016-11-24
Virtual simulation is a powerful technique for improving different types of systems, but the challenge is that the systems being simulated often behave in a stochastic way. This means that multiple simulations under the same conditions lead to different results. Florian Siegmund is a PhD student at the School of Engineering Science at the University of Skövde and on 12th of December he presents his doctoral thesis on multi-criteria optimization.
Florian has examined how optimization results can be improved by the use of an effective technique for repeated simulations. When connecting an automatic optimization to the simulation it becomes difficult for the optimization to assess which solutions really are the best and as a result the optimization result deteriorates. The title of the thesis is "Dynamic Resampling for Preference-Based Evolutionary Multi-Objective Optimization of Stochastic Systems".
– This technology simulates a particular configuration of the system several times and calculates the mean and variance. The repeated simulation reduces the uncertainty and helps to improve the optimization results, but it also has a price. Since the simulations are time consuming you can only perform a limited number of them, and when many simulations are used to evaluate a single configuration not as many different configurations can be evaluated, which is a disadvantage, says Florian.
Therefore, in his research, Florian develops new dynamic strategies for repeated simulation that distribute the number of simulations among the various configurations in a smart way. He uses evolutionary optimization techniques that primarily focus on the configurations that are of interest to the user. In this way, the optimization results become better and more relevant to the user and thus are a good tool for decision support. This is something that, for example, can be used to make car production, logistics, or patient flows in health care more effective.
– It's all about finding the best solutions for the various processes and within a limited time, perhaps only within a few hours, which has been difficult in the past. That is what I hope to achieve through the techniques I developed in my research. I have already tested them successfully on simulated systems at Volvo.
Florian Siegmund is from Germany and studied Computer Science at the University of Karlsruhe. One of his subjects was Operations Research, which is also part of the research at the University of Skövde and which thus forms the basis for his doctoral thesis.– It's about an interface between Computer Science, Economics and Engineering.
Before becoming a doctoral student at the university, Florian was an exchange student in Skövde and it was here that he wrote a student thesis. He also managed to study in Singapore for one semester before his PhD job became reality in Skövde. It was not just the research and doctoral hat that attracted him, Florian had also met his wife in Skaraborg.
Whether it will be Sweden or Germany in the future is not clear. Florian considers doing a so-called post-doc at the University of Skövde, but he has also applied for jobs in Gothenburg and in his home country.– My wish and hope is that I might continue to work in research and development.
Florian Siegmund presents his dissertation at the University of Skövde on Monday 12 February, in Portalen, Insikten, at 13:00.
Main supervisor: Professor Kalyanmoy Deb, Michigan State University, USA.
Supervisor: Professor Amos Ng, Högskolan i Skövde.
