For the last five years I have been working as a researcher and consultant for Daimler Benz AG, Research and Technology, Autonomous and Intelligent Systems group, Berlin, Germany. Till June 1998 I was also a Ph.D.-student in the Dynamics and Simulation of ecological Systems Group of the Department of Computer Science and Automation at the Technical University Ilmenau, Germany. At the beginning of my Ph.D. I worked as a visiting researcher in the Automatic Control and Systems Engineering group at the University of Sheffield, Great Britain.
During all these years I worked on the development and application of optimization techniques, especially evolutionary algorithms. This includes the development and test of optimization algorithms, the implementation of these algorithms for engineering use and their application to real world problems.
In my Ph.D. thesis "Development and Engineering Application of Evolutionary Algorithms" I present a number of extensions for evolutionary algorithms:
- An hierarchical structure of the procedures and operators of evolutionary algorithms. This facilitates a better understanding of the components as part of a larger system and the interactions between these parts.
- New extensions of the regional model: the use of multiple strategies and the use of competing subpopulations. Both are a further step for the development of efficient and enhanced evolutionary algorithms necessary for the solution of large and difficult problems
- Extensive methods for the visual analysis of evolutionary algorithms.
A longer and more complete description of the contents of the thesis is given in "Summary of the Ph.D. thesis".
During my work with evolutionary algorithms I used and implemented a number of different algorithms. This includes algorithms for parameter optimization (using different representations - real, integer, binary) and ordering problems and their corresponding operators in multiple variants. All these algorithms are fully integrated into one toolbox for consistent use (Genetic and Evolutionary Algorithm for use with Matlab). Additionally, the toolbox contains extensive visualization and reporting tools to analyze and examine the state and course of the optimization runs.
A number of real world applications was solved in collaboration with engineers and researchers from other research labs or groups. My part of the work contained the application of optimization methods to the problems and the development of problem specific analysis algorithms and methods and the extension of the given simulation programs and models.
- Optimization of the parameters of a combustion model of a diesel engine (concurrent fitting of multiple data sets, chaotic behavior of the large simulation program), together with researchers of Mercedes-Benz, engine development. This problem could be solved for the first time.
- Optimization of the parameters of a chopper controller (concurrent optimization of multiple scenarios), together with researchers of Daimler Benz, railway systems. Better solutions were found compared with previous results.
- Optimization of run-time of real time software modules (large scale systems), together with researchers of DaimlerChrysler, software testing group. Compared to systematic testing methods similar results could be found in a much shorter time-scale.
- Use of genetic programming for the design of models of dynamic systems and the design of controllers for dynamic systems, together with TU Darmstadt, Automatic Control group of Prof. Tolle.
- Optimization of the control of a coupled greenhouse-plant model using real world weather conditions (large scale dynamic system), together with Dr. Heißner from the Institute for Vegetable Production, Großbeeren/Berlin. First-time optimization of a fully integrated greenhouse model over a whole growing period.
Most of these problems were solved using evolutionary algorithms and the GEA Toolbox. Sometimes I also used deterministic optimization techniques (mostly gradient based), depending on the problem to solve.
During the work on the these problems both sides, the problem specialists and me, profited from the work and the results. Beside the solution of the problems every time previously unknown information about the problems could be found or derived and existing limitations were resolved. I could extend my optimization methods and tools and learn more about new application areas.
In my future work I would like to apply my experience in the area of optimization to new application areas and challenging real world problems and to extend my knowledge of optimization and other machine learning techniques.