Automatic Differentiation Applied for Optimization of Dynamical Systems

Petre Enciu (1) , Laurent Gerbaud (1) , Frédéric Wurtz (1)

1 G2ELab - Laboratoire de Génie Electrique de Grenoble

Abstract—Simulation is ubiquitous in many scientific areas. Applied for dynamic systems usually by employing differential equations, it gives the time evolution of system states. In order to solve such problems, numerical integration algorithms are often required. Automatic Differentiation (AD) is introduced as a powerful technique to compute derivatives of functions given in the form of computer programs in a high level programming lan- guage such as FORTRAN, C or C++. Such technique fits perfectly in combination with gradient based optimization algorithms, provided that the derivatives are valued with no truncation or cancellation error. This paper intends to use Automatic Differentiation employed for numerical integration schemes of dynamical systems simulating electromechanical actuators. Then, the resulting derivatives are used for sizing such devices by means of gradient based constrained optimization.

Domains: Electric power

Petre Enciu, Laurent Gerbaud, Frédéric Wurtz. Automatic Differentiation Applied for Optimization of Dynamical Systems. COMPUMAG 09, 17th Conference on the Computation of Electromagnetic Fields, Nov 2009, Florianopolis, Brazil. pp.Pages 249-250. ⟨hal-00439629⟩