Corrective Source Term Approach (CoTA) - Residual correction framework

Résumé fourni par la source

A software accompanying the paper "Correcting Misspecified Advection–Diffusion–Reaction Models under Chaotic Forcing: A Data-Driven Corrective Term Approach". In essence, we study how an imperfectphysics-based model (PBM) of a chaotically forced PDE can be corrected from temporally sparse data viaresidual correction mechanism, and which observables serve as informative regressors. The test system of interest is a one-dimensional advection–diffusion–reaction equation driven by aLorenz '05 type II velocity field, with model error injected into the transport and sourceterms. The method "Corrective Term Approach (CoTA)" has a data-driven model supplya state-space correction that is decoupled from the integrator, so the sampling interval maygreatly exceed the integration step. The model prediction at next sampling step is obtained by first integrating PBM over one sampling step,then a data-driven model (DDM) adds a learned corrective term to the resulting state. The network can beconditioned on lagged windows of the state, the velocity field, and the PBM's own one-step forecast. The software framework is modular in both halves of that pairing. The PBM and the DDM sit behindwrapper interfaces, so either can be replaced independently, i.e., a different PDE solver, adifferent equation/physical principle, a different DDM, all without modifying the hybrid model that combinesthem, enabling the study/usage of CoTA in various applications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.