Physics-Information Guide Non-linear Dynamics Identification
LI Yuanlu
WANG Jingyuan
WANG Yu
WU Zhiwei
Abstract：The structure of nonlinear systems is complex,and the current common identification methods have high require-ments for measurement data and poor noise robustness.How to select high accuracy and efficiency identification models for different nonlinear systems has been a huge challenge in the field of system identification.At present,the breakthrough of deep learning in data-driven and scientific computing has been applied to the field of dynamics,but the black box principle of deep learning model leads to its lack of interpretability.Based on this,a method of parameter identification by physic-informed neural network is pro-posed.By combining physical constraints with data,a loss function is constructed to make the network model follow the physical con-straints.The parameters to be identified are added to the network training process,will updated with the network training,and the weights that change at any time are added to optimize the training process.The results show that the system parameters can be accu-rately identified and the system state can be estimated with less data.Compared with traditional neural network models and data driv-en algorithms such as sparse identification,the method requires less data.The potential of feedback control as a state estimator is al-so demonstrated.
Keywords：physic informationneural networkdata-driven methodnonlinear dynamics
Publication Date：2025-08-20
Online Publishing Date：2025-12-12(First online date of this platform, not the publication date of the document)
Pages：6( 2156-2161 )