Automated reverse engineering of nonlinear dynamical systems

• DOI: 10.1073/pnas.0609476104
• Publication Date: 2007-06-07T00:54:51Z
• Authors: Josh Bongard; Hod Lipson
• Keywords: Dynamical system (definition); Physical system; Reverse engineering; Complex system

Complex nonlinear dynamics arise in many fields of science and engineering, but uncovering the underlying differential equations directly from observations poses a challenging task. The ability to symbolically model complex networked systems is key understanding them, an open problem disciplines. Here we introduce for first time method that can automatically generate symbolic coupled dynamical system series data. This applicable any be described using sets ordinary equations, assumes (possibly noisy) all variables are observable. Previous automated modeling approaches physical produced linear models or required provided manually. advance presented here made possible by allowing each coupled) variable separately, intelligently perturbing destabilizing extract its less observable characteristics, simplifying during modeling. We demonstrate this on four simulated two real spanning mechanics, ecology, biology. Unlike numerical models, have explanatory value, suggesting "reverse engineering" model-free identification may play increasing role our understand progressively more future.