Predicting Critical Transitions in Multiscale Dynamical Systems Using Reservoir Computing

Abstract

We study the problem of predicting rare critical transition events for a class of slow-fast nonlinear dynamical systems. The state of the system of interest is described by a slow process, whereas a faster process drives its evolution and induces critical transitions. By taking advantage of recent advances in reservoir computing, we present a data-driven method to predict the future evolution of the state. We show that our method is capable of predicting a critical transition event at least several numerical time steps in advance. We demonstrate the success as well as the limitations of our method using numerical experiments on three examples of systems, ranging from low dimensional to high dimensional. We discuss the mathematical and broader implications of our results.

arXiv comment21 pages

References in corpus (31)

• Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) Network
• Machine learning and the physical sciences
• Recent Advances in Physical Reservoir Computing: A Review
• Deep learning for universal linear embeddings of nonlinear dynamics
• Active Brownian Particles. From Individual to Collective Stochastic Dynamics
• Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations
• A high-bias, low-variance introduction to Machine Learning for physicists
• Tipping points in open systems: bifurcation, noise-induced and rate-dependent examples in the climate system
• Using Machine Learning to Replicate Chaotic Attractors and Calculate Lyapunov Exponents from Data
• Stochastic Parameterization: Towards a new view of Weather and Climate Models
• A mathematical framework for critical transitions: bifurcations, fast-slow systems and stochastic dynamics
• Attractor Reconstruction by Machine Learning
• Data-Driven Forecasting of High-Dimensional Chaotic Systems with Long Short-Term Memory Networks
• Hybrid Forecasting of Chaotic Processes: Using Machine Learning in Conjunction with a Knowledge-Based Model
• Nonlinear threshold behavior during the loss of Arctic sea ice
• Data-driven prediction of a multi-scale Lorenz 96 chaotic system using deep learning methods: Reservoir computing, ANN, and RNN-LSTM
• Embedding and Approximation Theorems for Echo State Networks
• Measurement of Anomalous Diffusion Using Recurrent Neural Networks
• Smooth approximation of stochastic differential equations
• A Note on Diffusion Limits of Chaotic Skew Product Flows
• A primer on noise-induced transitions in applied dynamical systems
• High-Performance Reconstruction of Microscopic Force Fields from Brownian Trajectories
• Using Synchronization for Prediction of High-Dimensional Chaotic Dynamics
• Homogenization for Deterministic Maps and Multiplicative Noise
• Data assimilation in slow-fast systems using homogenized climate models
• Deterministic Brownian Motion: The Effects of Perturbing a Dynamical System by a Chaotic Semi-Dynamical System
• Learn to Synchronize, Synchronize to Learn
• Bayesian optimization of hyper-parameters in reservoir computing
• Precursors to Rare Events in Stochastic Resonance
• Dynamical Systems as Temporal Feature Spaces
• Understanding Recurrent Neural Networks Using Nonequilibrium Response Theory

Cited by in corpus (4)

• Using Machine Learning to Anticipate Tipping Points and Extrapolate to Post-Tipping Dynamics of Non-Stationary Dynamical Systems
• Early warning signals for critical transitions in complex systems
• Time-series-analysis-based detection of critical transitions in real-world non-autonomous systems
• Neural Ordinary Differential Equations for Learning and Extrapolating System Dynamics Across Bifurcations