GEPROG - Genetic programming for modeling and control of dynamic systems

   Genetic Programming (GP) was originally developed to generate programs by imitating the natural evolution, that solve a given task. As programming languages initially small subsets were selected from languages that easily machined valid programs could be generated. This was necessary so that the variation of intermediate solutions simulated by mutation and recombination generated not strong structural breaks.

   The project GEPROG was to design a system based on genetic programming and evolution strategies method that generates for a given problem area, the modeling of nonlinear dynamic systems and the controller synthesis for non-linear plants, solutions that delivers real representation. These solutions should be generated in a form interpretable by the user, so that this post-process the results and can feed back into the optimization process.

   The target language in the project resulting tool are block diagrams for the simulation system MATLAB/SIMULINK. To ensure a high degree of interpretability, a two-stage evolutionary process has been designed in which the proposed structure variants of a GP algorithm are re-optimized using a parameter optimization on the basis of evolution strategy. This was the principle of strong causality (small changes to the appearance of a solution with high probability cause only small changes in its quality) implemented, which makes the optimization of direct search method in the first sense.

   In the project a complete tool for this purpose has been developed and successfully tested in industrial tasks. To apply the method is particularly useful for the data-driven modeling, if a non-linear relationship can be assumed and during the generation of controllers for systems with complicated or simply unknown dynamics.

 

Project partner:

  • Lohnert, F.; Schütte, A.; Sprave, J. (DaimlerChrysler AG, Forschung und Technologie FT3/AI)
  • Rechenberg, I.; Boblan, I.; Raab, U.; Santibáñez Koref, I. (Technische Universität Berlin, Fachgebiet Bionik und Evolutionstechnik)
  • Banzhaf, W.; Keller, R.; Niehaus, J.; Rauhe, H. (Informatik Centrum Dortmund e.V.)

Videos related to our lab

Citations related to our lab

Follow us to the Bionik-Kompetenz-Netz