A Dynamic Mode Decomposition Approach to Parameter Identification

arXiv:2604.18783v3 Announce Type: replace-cross Abstract: This paper develops a data-driven algorithm for simultaneous system and parameter identification in control-affine nonlinear systems. Parameter identification is achieved by training a data-driven predictive model using measurements collected across a prescribed set of known parameter values. The predictive model is then used to estimate unknown parameter values from measured trajectories by minimizing the trajectory prediction error using multi-start Nelder-Mead. Numerical experiments on the controlled Duffing oscillator with unknown damping, stiffness, and nonlinearity coefficients demonstrate accurate recovery of both the system trajectories and the unknown parameter values from data collected under open-loop excitation. Additional studies evaluate parameter recovery across randomly selected parameter values and the effects of training-set size, query-horizon length, measurement noise, and process noise.