Recurrent Neural Network Architecture Search for Geophysical Emulation

Maulik, Romit; Égelé, Romain; Lusch, Bethany; Balaprakash, Prasanna

doi:10.1109/sc41405.2020.00012

Cited by 36 publications

(34 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The framework shows deviations in the finer scale content (mode 3) as one approaches the end of the testing period, indicating potential extrapolation. Similar behaviour was observed in [33,35] as well, where a multi-cell LSTM was used to forecast in this reduced space (henceforth POD-LSTM). Time-series assessments for various point probes in the Eastern Pacific are shown for a testing sub-window where data from all prediction sources were available (between 5 April 2015 and 17 June 2018) in figure 4.…”

Section: Methodssupporting

confidence: 80%

“…Some qualitative comparisons of the predictions are shown in figure 5, where an acceptable agreement between different methods and the remote-sensing dataset is observed (table 1). We also note that the metrics obtained using the optimized LSTM architecture in this table outperformed the accuracy obtained from classical linear or decision-tree-based forecasting techniques (see table 2 in [33]). A closer examination of the mean squared error during the entire testing period is shown in figure 6, where POD-RKHS is seen to give sufficiently accurate predictions in the entire domain, including the vital Eastern Pacific region.…”

Section: Methodsmentioning

confidence: 72%

“…Although the choice of user-defined hyperparameters in both these methods may change these values, their training times are comparable. We note that, in a manner similar to [33], discovering the optimal hyperparameters and architectures for a deep learning framework adds significant costs associated with sampling a high-dimensional search space of neural networks that may fit the given training data. In that regard, kernel methods may provide significant computational gains.…”

Section: Methodsmentioning

confidence: 99%

“…Further opportunities and perspectives for the use of data-driven methods for geosciences may be found in the reviews of [32,36]. Before proceeding, we note that a vast majority of the data-driven developments for geophysical forecasting have involved the use of variants of deep learning methods, for example ResNets [37], CapsuleNets [22], U-Nets [31], long short-term memory networks (LSTMs) [33], convolutional-LSTMs [34], neural ordinary differential equations (ODEs) [38,39] and local or global fully connected deep neural networks [32]. These methods, while exceptionally powerful in learning complex functions, hamper interpretability and require large computational resources for optimization.…”

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

Simple, low-cost and accurate data-driven geophysical forecasting with learned kernels

2021

View full text Add to dashboard Cite

Modelling geophysical processes as low-dimensional dynamical systems and regressing their vector field from data is a promising approach for learning emulators of such systems. We show that when the kernel of these emulators is also learned from data (using kernel flows, a variant of cross-validation), then the resulting data-driven models are not only faster than equation-based models but are easier to train than neural networks such as the long short-term memory neural network. In addition, they are also more accurate and predictive than the latter. When trained on geophysical observational data, for example the weekly averaged global sea-surface temperature, considerable gains are also observed by the proposed technique in comparison with classical partial differential equation-based models in terms of forecast computational cost and accuracy. When trained on publicly available re-analysis data for the daily temperature of the North American continent, we see significant improvements over classical baselines such as climatology and persistence-based forecast techniques. Although our experiments concern specific examples, the proposed approach is general, and our results support the viability of kernel methods (with learned kernels) for interpretable and computationally efficient geophysical forecasting for a large diversity of processes.

show abstract

Section: Methodssupporting

confidence: 80%

Section: Methodsmentioning

confidence: 72%

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Simple, low-cost and accurate data-driven geophysical forecasting with learned kernels

2021

View full text Add to dashboard Cite

show abstract

“…In addition to enriching the input feature space as explained above, we can improve the trustworthiness of data‐driven parts by imposing constraints and physical laws on the methods through the regularization of the cost function [276] as well as by developing a scalable and optimal neural architecture search [212] approach for model structure selection. To develop a more quantitative insight into the effects of these approaches, we can exploit the use of piece‐wise affine representation of networks as shown in our recent work [285].…”

Section: Hybrid Analysis and Modelingmentioning

confidence: 99%

Hybrid analysis and modeling, eclecticism, and multifidelity computing toward digital twin revolution

2021

View full text Add to dashboard Cite

Most modeling approaches lie in either of the two categories: physics‐based or data‐driven. Recently, a third approach which is a combination of these deterministic and statistical models is emerging for scientific applications. To leverage these developments, our aim in this perspective paper is centered around exploring numerous principle concepts to address the challenges of (i) trustworthiness and generalizability in developing data‐driven models to shed light on understanding the fundamental trade‐offs in their accuracy and efficiency and (ii) seamless integration of interface learning and multifidelity coupling approaches that transfer and represent information between different entities, particularly when different scales are governed by different physics, each operating on a different level of abstraction. Addressing these challenges could enable the revolution of digital twin technologies for scientific and engineering applications.

show abstract

Predicting wind farm operations with machine learning and the P2D‐RANS model: A case study for an AWAKEN site

Moss,

Maulik,

Moriarty

et al. 2023

Wind Energy

View full text Add to dashboard Cite

The power performance and the wind velocity field of an onshore wind farm are predicted with machine learning models and the pseudo‐2D RANS model, then assessed against SCADA data. The wind farm under investigation is one of the sites involved with the American WAKE experimeNt (AWAKEN). The performed simulations enable predictions of the power capture at the farm and turbine levels while providing insights into the effects on power capture associated with wake interactions that operating upstream turbines induce, as well as the variability caused by atmospheric stability. The machine learning models show improved accuracy compared to the pseudo‐2D RANS model in the predictions of turbine power capture and farm power capture with roughly half the normalized error. The machine learning models also entail lower computational costs upon training. Further, the machine learning models provide predictions of the wind turbulence intensity at the turbine level for different wind and atmospheric conditions with very good accuracy, which is difficult to achieve through RANS modeling. Additionally, farm‐to‐farm interactions are noted, with adverse impacts on power predictions from both models.

show abstract

Recurrent Neural Network Architecture Search for Geophysical Emulation

Cited by 36 publications

References 30 publications

Simple, low-cost and accurate data-driven geophysical forecasting with learned kernels

Simple, low-cost and accurate data-driven geophysical forecasting with learned kernels

Hybrid analysis and modeling, eclecticism, and multifidelity computing toward digital twin revolution

Predicting wind farm operations with machine learning and the P2D‐RANS model: A case study for an AWAKEN site

Contact Info

Product

Resources

About