Hybrid modeling of a multidimensional coupled nonlinear system with integration of Hamiltonian mechanics

• DOI: 10.1007/s11071-023-08618-0
• Publication Date: 2023-06-27T05:01:32Z
• Authors: Amirhassan Abbasi; Prashant N. Kambali; C. Nataraj
• Keywords: Hamiltonian (control theory)

Abstract This study concerns hybrid modeling of a multidimensional coupled nonlinear system. The underlying basis for the model is derived from Hamiltonian mechanics capitalizing on broad utility and efficiency energy-based reasoning in high-dimensional systems. essentially an artificial neural network with computational graph that modified conventional networks few significant ways. first modification includes incorporating intermediate scalar function representing learned data. second enhances input/output channels capturing dynamics main goal such to improve extrapolation capability by enforcing conformance some key aspects physics form bias. results demonstrate this physics-based bias into empowers it produce long-term physically plausible predictions. proposed approach also shows high scalability dynamic systems general.