Registered Data

[00790] Machine learning methods with error analysis for optimal control problems

  • Session Time & Room : 5D (Aug.25, 15:30-17:10) @F312
  • Type : Contributed Talk
  • Abstract : We consider optimal control with partial differential equations (()PDE()) and present a numerical method based on machine learning including control error analysis. Physics-Informed Neural Networks (()PINN()) are used with the cost and penalty terms for the PDE as loss function. The model size is iteratively increased until the a posteriori estimated control error satisfies a given accuracy. The method is illustrated with numerical examples for 1D heat transfer and 3D turbine activation.
  • Classification : 49M41, 49M25, 68T05, 65G20
  • Format : Talk at Waseda University
  • Author(s) :
    • Georg Vossen (Kreleld University of Applied Sciences)
    • Semih Sirin (Kreleld University of Applied Sciences)
    • Nicolai Friedlich (Kreleld University of Applied Sciences)