Learning dynamical systems using local stability priors

A computational approach to simultaneously learn the vector field of a dynamical system with locally asymptotically stable equilibrium and its region attraction from system's trajectories is proposed. The nonlinear identification leverages local stability information as prior on system, effectively endowing estimate this important structural property. In addition, knowledge can be used design experiments by informing selection initial conditions which are generated enabling use Lyapunov function regularization term. Simulation results show that proposed method allows efficient sampling provides an accurate dynamics in inner approximation attraction.