A few-shot identification method for stochastic dynamical systems based on residual multipeaks adaptive sampling
Top Cited Papers
- 1 July 2024
- journal article
- research article
- Published by AIP Publishing in Chaos: An Interdisciplinary Journal of Nonlinear Science
- Vol. 34 (7)
Abstract
Neural networks are popular data-driven modeling tools that come with high data collection costs. This paper proposes a residual-based multipeaks adaptive sampling (RMAS) algorithm, which can reduce the demand for a large number of samples in the identification of stochastic dynamical systems. Compared to classical residual-based sampling algorithms, the RMAS algorithm achieves higher system identification accuracy without relying on any hyperparameters. Subsequently, combining the RMAS algorithm and neural network, a few-shot identification (FSI) method for stochastic dynamical systems is proposed, which is applied to the identification of a vegetation biomass change model and the Rayleigh–Van der Pol impact vibration model. We show that the RMAS algorithm modifies residual-based sampling algorithms and, in particular, reduces the system identification error by 76% with the same sample sizes. Moreover, the surrogate model accurately predicts the first escape probability density function and the P bifurcation behavior in the systems, with the error of less than 1.59×10−2. Finally, the robustness of the FSI method is validated.Keywords
Funding Information
- National Natural Science Foundation of China (12172291)
- National Natural Science Foundation of China (12232015)
- the 111 Project (BP0719007)
- Shaanxi Province Outstanding Youth Fund Project (2024JC-JCQN-05)
This publication has 27 references indexed in Scilit:
- A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networksComputer Methods in Applied Mechanics and Engineering, 2022
- Stochastic physics-informed neural ordinary differential equationsJournal of Computational Physics, 2022
- Residual-based adaptivity for two-phase flow simulation in porous media using Physics-informed Neural NetworksComputer Methods in Applied Mechanics and Engineering, 2022
- Adaptive deep neural networks methods for high-dimensional partial differential equationsJournal of Computational Physics, 2022
- The stochastic P-bifurcation analysis of the impact system via the most probable responseChaos, Solitons & Fractals, 2021
- Uncertain inverse problem for fractional dynamical systems using perturbed collage theoremCommunications in Nonlinear Science and Numerical Simulation, 2021
- Noise-induced vegetation transitions in the Grazing EcosystemApplied Mathematical Modelling, 2019
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equationsJournal of Computational Physics, 2019
- Parametric and nonparametric system identification of an experimental turbojet engineAerospace Science and Technology, 2015
- Demonstration and validation of constructive initialization method for neural networks to approximate nonlinear functions in engineering mechanics applicationsNonlinear Dynamics, 2014