PT - JOURNAL ARTICLE
AU - Qi, Di
AU - Majda, Andrew J.
TI - Using machine learning to predict extreme events in complex systems
AID - 10.1073/pnas.1917285117
DP - 2020 Jan 07
TA - Proceedings of the National Academy of Sciences
PG - 52--59
VI - 117
IP - 1
4099 - http://www.pnas.org/content/117/1/52.short
4100 - http://www.pnas.org/content/117/1/52.full
SO - Proc Natl Acad Sci USA2020 Jan 07; 117
AB - Understanding and predicting extreme events as well as the related anomalous statistics is a grand challenge in complex natural systems. Deep convolutional neural networks provide a useful tool to learn the essential model dynamics directly from data. A deep learning strategy is proposed to predict the extreme events that appear in turbulent dynamical systems. A truncated KdV model displaying distinctive statistics from near-Gaussian to highly skewed distributions is used as the test model. The neural network is trained using data only from the near-Gaussian regime without the occurrence of large extreme values. The optimized network demonstrates uniformly high skill in successfully capturing the solution structures in a wide variety of statistical regimes, including the highly skewed extreme events.Extreme events and the related anomalous statistics are ubiquitously observed in many natural systems, and the development of efficient methods to understand and accurately predict such representative features remains a grand challenge. Here, we investigate the skill of deep learning strategies in the prediction of extreme events in complex turbulent dynamical systems. Deep neural networks have been successfully applied to many imaging processing problems involving big data, and have recently shown potential for the study of dynamical systems. We propose to use a densely connected mixed-scale network model to capture the extreme events appearing in a truncated Kortewegâ€“de Vries (tKdV) statistical framework, which creates anomalous skewed distributions consistent with recent laboratory experiments for shallow water waves across an abrupt depth change, where a remarkable statistical phase transition is generated by varying the inverse temperature parameter in the corresponding Gibbs invariant measures. The neural network is trained using data without knowing the explicit model dynamics, and the training data are only drawn from the near-Gaussian regime of the tKdV model solutions without the occurrence of large extreme values. A relative entropy loss function, together with empirical partition functions, is proposed for measuring the accuracy of the network output where the dominant structures in the turbulent field are emphasized. The optimized network is shown to gain uniformly high skill in accurately predicting the solutions in a wide variety of statistical regimes, including highly skewed extreme events. The technique is promising to be further applied to other complicated high-dimensional systems.