A DREM-Based Approach for the Identification of Chaotic Systems

A straightforward methodology for identifying certain classes of chaotic systems based on a novel version of the least-squares method, assuming they are algebraically observable and identifiable with respect to a measurable output, is introduced. This output allows us to express the original system as a chain of integrators, where the last term, which depends on the output and its corresponding time derivatives, lumps the system's non-linearities. We can factorize this term into a regressor function multiplied by an unknown-parameter vector, suggesting that a high-gain observer can be used to simultaneously and approximately estimate the states of the pure integrator and the evolution of the lumped nonlinear term. This allows us to rewrite the original system as a linear regression equation. This configuration enables the above-mentioned least-squares method to recover the chaotic-system parameters.