Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Datasets”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,783 records · Page 99

On the feasibility of using physics-informed machine learning for underground reservoir pressure management

In this work, we evaluate the feasibility of using physics-informed machine learning (PIML) for underground energy-related pressure management. To this end, we develop a PIML framework to manage underground reservoir pressures by training neural networks to determine fluid extraction rates for dedicated extraction wells during fluid injection operations given a range of reservoir conditions (e.g., transmissivity and storativity). We implement an automatically-differentiable analytical physics model of fluid flow in porous media within the PIML framework as a proxy for more complicated models. This allows us to execute a sufficient number of training scenarios to fully evaluate the feasibility of using PIML to support pressure management activities. We quantify the number of physics-model parameters required for automatic differentiation to become more efficient than finite-difference gradient calculations. We use a simple scenario with a single injector, extractor, and critical location for our feasibility analysis. We evaluate the effect of the size of the training dataset (i.e., the number of reservoir condition samples) on the accuracy and efficiency of the PIML framework. For an equivalent number of model evaluations, the larger training dataset took less time to train and produced a neural network that was able to more accurately manage reservoir pressures. We also evaluate the effect of the training dataset batch size (i.e., number of reservoir condition samples used to update the neural network coefficients during training; i.e., how the training dataset is partitioned). While training ran faster with larger batch sizes, they produced neural networks that managed pressures less accurately. We demonstrate the approach on a more complex scenario involving 10 injectors, 10 extractors, and 4 critical locations (a relatively high well density of 20 wells/km2). We provide the number of forward and adjoint model evaluations required in each case as an indication of the feasibility of using PIML for pressure management when more complicated physics models with longer execution times are used.

54 ENVIRONMENTAL SCIENCES↗

Common Column Identification for Table Similarity Detection in Electrified Transportation Data Lakes

Electrified transportation often requires researchers and operators to interact with datasets from a wide range of sources and disciplines, such as transportation, power systems, public health, policies, and regulations. These datasets vary in quality and format, making it difficult to understand, preprocess, and identify key columns representing real-world entities or values for indexing and joining, which can negatively impact downstream analysis and operation. Existing solutions are limited, requiring extensive manual customization or data expertise to utilize. In this article, we propose a multi-layered approach to automatically identify key columns to expedite preprocessing and aid in analysis of electrified transportation data. Our method leverages a dynamic ontology to identify common fields and an information theory-based strategy for edge cases that are difficult to generalize. Evaluations on a number of datasets from data.gov and kaggle.com show improved performance of our methods over several baseline techniques, and our ablation analyses illustrate the efficacy of individual components of our method. Our case studies also demonstrate that our methods have the potential to improve analysis of electrified transportation data and aid in automatic integration of such datasets.

33 ADVANCED PROPULSION SYSTEMS↗

An Anatomically Validated Brachial Plexus Contouring Method for Intensity Modulated Radiation Therapy Planning

Purpose: To develop contouring guidelines for the brachial plexus (BP) using anatomically validated cadaver datasets. Magnetic resonance imaging (MRI) and computed tomography (CT) were used to obtain detailed visualizations of the BP region, with the goal of achieving maximal inclusion of the actual BP in a small contoured volume while also accommodating for anatomic variations. Methods and Materials: CT and MRI were obtained for 8 cadavers positioned for intensity modulated radiation therapy. 3-dimensional reconstructions of soft tissue (from MRI) and bone (from CT) were combined to create 8 separate enhanced CT project files. Dissection of the corresponding cadavers anatomically validated the reconstructions created. Seven enhanced CT project files were then automatically fitted, separately in different regions, to obtain a single dataset of superimposed BP regions that incorporated anatomic variations. From this dataset, improved BP contouring guidelines were developed. These guidelines were then applied to the 7 original CT project files and also to 1 additional file, left out from the superimposing procedure. The percentage of BP inclusion was compared with the published guidelines. Results: The anatomic validation procedure showed a high level of conformity for the BP regions examined between the 3-dimensional reconstructions generated and the dissected counterparts. Accurate and detailed BP contouring guidelines were developed, which provided corresponding guidance for each level in a clinical dataset. An average margin of 4.7 mm around the anatomically validated BP contour is sufficient to accommodate for anatomic variations. Using the new guidelines, 100% inclusion of the BP was achieved, compared with a mean inclusion of 37.75% when published guidelines were applied. Conclusion: Improved guidelines for BP delineation were developed using combined MRI and CT imaging with validation by anatomic dissection.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

A unified approach to evaluating precipitation frequency estimates with uncertainty quantification: Application to Florida and California watersheds

Although a variety of climate datasets are now available, including observational and model products, there remains a pressing need for the development of a unified, automated and application-focused approach for evaluating and comparing climate extremes within and across these datasets. Precipitation frequency (PF) es- timates for extreme precipitation events are useful decision-relevant quantities among water managers. Our objective and quantitative framework for computing PF estimates of any duration uses known techniques in a novel combination. Our framework consists of three steps: (1) use of a regionalization-based statistical method for precipitation frequency estimation at observed sites; this involves hierarchical clustering (here we used Ward’s method) for delineating homogeneous regions, and regional frequency analysis using L-moments; (2) use of kriging for spatial mapping because of its advantages over other conventional interpolation methods; (3) use of z-scores accounts for estimation uncertainties when comparing PF estimates between datasets. The 24-h PF estimates in the Kissimmee-Southern Florida watershed (flat topography) and the Sacramento-San Joaquin watershed (complex topography) demonstrate the applicability of this approach in regions of any geographical complexity. Our results are shown to be reasonably calibrated using the probability integral transform (PIT) histogram of the best fitted distribution. Our results are compared with that of the NOAA Atlas 14 reports and show that for most of the stations the two estimates are statistically indistinguishable; and, the confidence intervals of our estimates are narrower than those of the NOAA estimates. Our approach is applicable to a variety of datasets and provides a baseline for assessing performance of climate models in historical simulations – a necessary first step towards analyzing future projections.

58 GEOSCIENCES↗

Predicting future well performance for environmental remediation design using deep learning

Here in this study, we developed a deep learning (DL) framework with a multi-channel three-dimensional convolutional neural network (MC3D-CNN) to predict well performance and thereby assist future environmental remediation design. Such prediction of extraction well performance at designated locations is critical for configuring pump-and-treat (P&T) well network design and operation, setting reasonable target closure dates for overall remedying, and estimating remedy costs. The framework is developed with operational and monitoring data routinely collected during P&T remedy operations, including well extraction and injection rates as well as in situ contaminant concentrations. Traditionally, the collected data were rarely used for purposes other than assessing past well performance and the accuracy of the conceptual site model. However, recent advances in data-driven computational approaches enable better use of the large datasets to inform future well performance, enhance site characterization, and improve remediation planning. In this study, we established a DL framework to integrate transient three-dimensional contaminant plumes and multiple aquifer properties (e.g., hydraulic conductivity and hydrostratigraphic maps) to identify characteristic patterns controlling and representing extraction well mass recovery, aiming at providing future mass recovery estimates for existing wells and candidate wells at any proposed locations. We evaluated our framework by using a realistic synthetic dataset generated from a well-calibrated flow and transport model used in the 200 West Area of the U.S. Department of Energy’s Hanford Site in southeastern Washington state. The multi-channel feature in our framework allows integration of various types and temporal densities of training datasets for DL model development. Overall, we found that the trained DL model achieved an accuracy of over 90% in ranking extraction well performance in validation datasets, and over 80% in predicting high-performance-ranking well locations. This data-informed approach provides a flexible tool to support adaptive site management, streamline decision-making, and potentially reduce remediation time and costs. Our DL framework can be used as a filtering tool to improve the current P&T network optimization design by reducing the number of candidate well locations.

54 ENVIRONMENTAL SCIENCES↗

Machine learning for the redox potential prediction of molecules in organic redox flow battery

Here, organic redox flow batteries (ORFB) are recognized as an innovative technology for the large-scale storage of renewable energy. The redox potential of organic redox-active molecules plays a vital role in their performance. Advanced screening techniques like high-throughput experiment and machine learning (ML) have significantly enhanced organic material performance and transformed the field of ORFB. However, the scarcity of experimental data poses a considerable challenge for ML model development in this domain. In our study, we developed lightweight graph-based Gaussian process regression (GPR) models with GPU-accelerated marginalized graph kernel and hybrid kernel to predict the redox potentials of organic redox-active molecules for ORFBs, specifically focusing on small datasets. To evaluate model accuracy, we created a new experimental database of organic redox-active molecules by the data from hundreds of published papers and assembled previous computational datasets. We also considered some key parameters, such as pH conditions and solvent type, to assess their impact on redox potential prediction. Our GPR model predicted redox potentials with high accuracy across all datasets using minimal training data. The study provides powerful tools for molecule screening and design and delivers valuable guidance on designing training datasets for costly experiments.

25 ENERGY STORAGE↗

Multimodality Image Fusion and Planning and Dose Delivery for Radiation Therapy

Image-guided radiation therapy (IGRT) relies on the quality of fused images to yield accurate and reproducible patient setup prior to dose delivery. The registration of 2 image datasets can be characterized as hardware-based or software-based image fusion. Hardware-based image fusion is performed by hybrid scanners that combine 2 distinct medical imaging modalities such as positron emission tomography (PET) and computed tomography (CT) into a single device. In hybrid scanners, the patient maintains the same position during both studies making the fusion of image data sets simple. However, it cannot perform temporal image registration where image datasets are acquired at different times. On the other hand, software-based image fusion technique can merge image datasets taken at different times or with different medical imaging modalities. Software-based image fusion can be performed either manually, using landmarks, or automatically. In the automatic image fusion method, the best fit is evaluated using mutual information coefficient. Manual image fusion is typically performed at dose planning and for patient setup prior to dose delivery for IGRT. The fusion of orthogonal live radiographic images taken prior to dose delivery to digitally reconstructed radiographs will be presented. Although manual image fusion has been routinely used, the use of fiducial markers has shortened the fusion time. Automated image fusion should be possible for IGRT because the image datasets are derived basically from the same imaging modality, resulting in further shortening the fusion time. The advantages and limitations of both hardware-based and software-based image fusion methodologies are discussed.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Investigation of Simple IMRT Delivery Techniques for Non-Small Cell Lung Cancer Patients with Respiratory Motion Using 4DCT

Techniques for generating simplified IMRT treatment plans for treating non-small cell lung cancer (NSCLC) patients with respiratory motion were investigated. To estimate and account for respiratory motion, 4-dimensional computed tomography (4DCT) datasets from 5 patients were used to design 5-field 6-MV ungated step-and-shoot intensity modulated radiotherapy (IMRT) plans delivering a dose of 66 Gy to the planning target volume (PTV). For each patient, 2 plans were generated using the mean intensity and the maximum intensity of 10 CT datasets from different breathing phases. The plans also utilized different margins around the clinical target volume/internal target volume (CTV/ITV) to account for tumor motion. To reduce the treatment time and ensure accurate dose delivery to moving targets, the number of intensity levels was minimized while maintaining dose coverage to PTV and minimizing dose to organs at risk (OARs). Dose-volume histograms (DVHs), dosimetric metrics, and outcome probabilities were evaluated for all plans. Plans using the averaged CT image dataset were inferior, requiring larger margins around the PTV, with a maximum of 1.5 cm, to ensure coverage of the tumor, and therefore increased the dose to OARs located in proximity of the tumor. The plans based on superimposed CT image datasets achieved full coverage of the tumor, while allowing tight margins around the PTV and minimizing the dose to OARs. A small number of intensity-levels (3 to 5), resulting in IMRT plans with a total of 13 to 30 segments, were sufficient for homogeneous PTV coverage, without affecting the sparing of OARs. In conclusion, a technique involving treatment planning with the superimposed CT scans of all respiratory phases, and the application of IMRT with only a small number of segments was feasible despite significant tumor motion; however, greater patient numbers are needed to support the statistical significance of the results presented in this work.

61 RADIATION PROTECTION AND DOSIMETRY↗

Machine learning prediction on the fractional free volume of polymer membranes

Fractional free volume (FFV) characterizes the microstructural level features of polymers and affects their properties including thermal, mechanical, and separation performance. Experimental measurements and theoretical analyses have been used to quantify the FFV of polymers, but challenges remain because of their limitations. Experimental measurements are laborious and based on semi empirical equations, while Bondi’s group contribution theory involves ambiguities like the determination of van der Waals volume and the choice of factor values in the theoretical equation. To efficiently evaluate the FFV of polymers, this study utilizes high-throughput molecular dynamics (MD) simulations to build a large dataset regarding polymer’s FFV. Based on this large dataset, we further build machine learning (ML) models to establish the composition-structure relation. Inspired by group contribution theory which correlates polymer’s functional groups to FFV, our ML models correlate polymer’s substructures or physico-chemical indexes to FFV. Here, our study first benchmarks the MD simulation protocol to obtain reliable FFV of polymers and then carries out high-throughput MD simulations for more than 6,500 homopolymers and 1,400 polyamides. Such a large and diverse dataset makes the well-trained ML models more generalizable, compared with the group contribution theory. The efficiency of a feed forward neural network model is further demonstrated by applying it to a hypothetical polyimide dataset of more than 8 million chemical structures. The predicted FFVs of hypothetical polyimides are further validated by MD simulations. The obtained FFVs of the 8 million polymers, plus their previously reported gas separation performances, demonstrate the promising capability of ML virtual screening for the discovery of polymer membranes with exceptional permeability/selectivity.

36 MATERIALS SCIENCE↗

Recommended Nuclear Data for the Production of Selected Therapeutic Radionuclides

Increasing interest in the treatment of human disease by means of targeted radionuclide-based therapies requires an accurate understanding of achievable yields and quantification of realisable degrees of purity. Important studies within a more extensive Coordinated Research Project of the International Atomic Energy Agency (IAEA CRP) have included in-depth assessments and evaluations of all available measured cross-section data for ten nuclear reactions leading to the formation of 131 Cs, 131 Ba, 225 Ac, 225 Ra, 227 Th, 227 Ac, 230 U and 230 Pa for various on-going and proposed therapeutic applications in nuclear medicine. Equivalent studies were also undertaken of possible reactions to produce the 178W /low-spin 178m Ta generator system that has been clinically applied in diagnosis by means of SPECT - these particular cross-section studies and recommendations are included in this paper for completeness. All available measured data have been critically reviewed and assessed to produce statistically consistent datasets. Selected datasets have been fitted by a least-squares method with Padé approximations of variable order to evaluate the production cross sections and corresponding uncertainties. New measurements have been made in some instances, and reported data have been adjusted to accommodate new nuclear decay or monitor reaction data which altered the original reported quantities adversely. Insufficient measured data are available in many cases to permit a reliable evaluation of cross sections, and these instances are discussed. Finally, recommended therapeutic radionuclide production reaction data and their uncertainties are available on the Web page of the IAEA Nuclear Data Section at www-nds.iaea.org/radionuclides/ and also at the IAEA medical portal www-nds.iaea.org/medportal/.

63 RADIATION, THERMAL, AND OTHER ENVIRON. POLLUTAN↗

Nuclear Data Sheets for A=50

The experimental nuclear structure and decay data are evaluated for all the known nuclides of mass 50 (Cl, Ar, K, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni). Here, detailed evaluated nuclear structure information are presented in individual decay and reaction datasets and the best values are recommended in the Adopted dataset for level properties, γ and β radiations, and other spectroscopic information. No excited states are known in 50Cl and 50Ni, and limited information is available for 50 Ar, 50 K, 50 Fe and 50 Co. This work supersedes earlier evaluations of A=50 by 2011El01, 1995Bu09, 1990Bu18, 1984Al29, and 1976Au07.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Persistent global greening over the last four decades using novel long-term vegetation index data with enhanced temporal consistency

Advanced Very High-Resolution Radiometer (AVHRR) satellite observations have provided the longest global daily records from 1980s, but the remaining temporal inconsistency in vegetation index datasets has hindered reliable assessment of vegetation greenness trends. To tackle this, we generated novel global long-term Normalized Difference Vegetation Index (NDVI) and Near-Infrared Reflectance of vegetation (NIRv) datasets derived from AVHRR and Moderate Resolution Imaging Spectroradiometer (MODIS). We addressed residual temporal inconsistency through three-step post processing including cross-sensor calibration among AVHRR sensors, orbital drifting correction for AVHRR sensors, and machine learning-based harmonization between AVHRR and MODIS. After applying each processing step, we confirmed the enhanced temporal consistency in terms of detrended anomaly, trend and interannual variability of NDVI and NIRv at calibration sites. Our refined NDVI and NIRv datasets showed a persistent global greening trend over the last four decades (NDVI: 0.0008 yr -1 ; NIRv: 0.0003 yr -1 ), contrasting with those without the three processing steps that showed rapid greening trends before 2000 (NDVI: 0.0017 yr -1 ; NIRv: 0.0008 yr -1 ) and weakened greening trends after 2000 (NDVI: 0.0004 yr -1 ; NIRv: 0.0001 yr -1 ). These findings highlight the importance of minimizing temporal inconsistency in long-term vegetation index datasets, which can support more reliable trend analysis in global vegetation response to climate changes.

54 ENVIRONMENTAL SCIENCES↗

Monthly average clear-sky broadband irradiance database for worldwide solar heat gain and building cooling load calculations

This paper establishes the formulation of a new clear-sky solar radiation model appropriate for algorithms calculating cooling loads in buildings. The aim is to replace the ASHRAE clear-sky model of 1967, whose limitations are well known and are reviewed. The new model is derived in two steps. The first step consists of obtaining a reference irradiance dataset from the REST2 model, which uses a high-performance, validated, two-band clear-sky algorithm. REST2 requires detailed inputs about atmospheric conditions such as aerosols, water vapor, ozone, and ground albedo. The development of global atmospheric datasets used as inputs to REST2 is reviewed. For the most part, these datasets are derived from space observations to guarantee universality and accuracy. In the case of aerosols, point-source terrestrial measurements were also used as ground truthing of the satellite data. The second step of the model consists of fits derived from a REST2-based reference irradiance dataset. These fits enable the derivation of compact, but relatively accurate expressions, for beam and diffuse clear-sky irradiance. The fitted expressions require the tabulation of only two pseudo-optical depths for each month of the year. The resulting model, and its tabulated data, are expected to be incorporated in the 2009 edition of the ASHRAE Handbook of Fundamentals. (author)

14 SOLAR ENERGY↗

Data integration reveals key homeostatic mechanisms following low dose radiation exposure

The goal of this study was to define pathways regulated by low dose radiation to understand how biological systems respond to subtle perturbations in their environment and prioritize pathways for human health assessment. Using an in vitro 3-D human full thickness skin model, we have examined the temporal response of dermal and epidermal layers to 10 cGy X-ray using transcriptomic, proteomic, phosphoproteomic and metabolomic platforms. Bioinformatics analysis of each dataset independently revealed potential signaling mechanisms affected by low dose radiation, and integrating data shed additional insight into the mechanisms regulating low dose responses in human tissue. We examined direct interactions among datasets (top down approach) and defined several hubs as significant regulators, including transcription factors (YY1, MYC and CREB1), kinases (CDK2, PLK1) and a protease (MMP2). These data indicate a shift in response across time — with an increase in DNA repair, tissue remodeling and repression of cell proliferation acutely (24–72 h). Pathway-based integration (bottom up approach) identified common molecular and pathway responses to low dose radiation, including oxidative stress, nitric oxide signaling and transcriptional regulation through the SP1 factor that would not have been identified by the individual data sets. Significant regulation of key downstream metabolites of nitrative stress was measured within these pathways. Among the features identified in our study, the regulation of MMP2 and SP1 was experimentally validated. Our results demonstrate the advantage of data integration to broadly define the pathways and networks that represent the mechanisms by which complex biological systems respond to perturbation. - Highlights: • Low dose ionizing radiation altered homeostasis in 3D skin tissue model. • Global gene/protein/metabolite data integrated using complementary statistical approaches • Time and location-specific change in matrix regulation, ROS/RNS and DNA repair pathways detected • Low dose exposure alters metabolites involved in nitric oxide biosynthesis and wound healing. • Computationally predicted regulators of primary mechanisms were experimentally validated.

60 APPLIED LIFE SCIENCES↗

Evaluation of historical CMIP6 model simulations of extreme precipitation over contiguous US regions

Simulated historical precipitation is evaluated for Coupled Model Intercomparison Project Phase 6 (CMIP6) models using precipitation indices defined by the Expert Team on Climate Change Detection and Indices. The model indices are evaluated against corresponding indices from the CPC unified gauge-based analyses of precipitation over seven geographical regions across the contiguous US (CONUS). The regions assessed match those in recent US National Climate Assessment Reports. To estimate observational uncertainty, precipitation indices for three other observational datasets (HadEx2, Livneh and PRISM) are evaluated against the CPC analyses. Both the moderate and extreme mean precipitation intensities are overestimated over the western CONUS and underestimated in the areas of the Central Great Plains (CGP) in most CMIP6 models tested. Most CMIP6 models overestimate the mean and variability of wet spell durations and underestimate the mean and variability of dry spell durations across the CONUS. Biases in interannual variability of most of the indices have similar patterns to those in corresponding mean biases. The median and interquartile model spreads in CMIP6 model biases are clearly smaller than those in CMIP5 model biases for wet spell durations. Multimodel medians of CMIP6 (CMIP6-MMM) and CMIP5 (CMIP5-MMM) have similar biases in climatology and variability but biases tend to be smaller in CMIP6-MMM. Depending on the index, extreme precipitation is slightly better in parts of the eastern half of the CONUS in CMIP6-MMM, otherwise, the biases in climatology and variability are similar to CMIP5-MMM. CMIP6-MMM performs better than individual models and even observational datasets in some cases. Differences between observational datasets for most indices are comparable to the CMIP6 interquartile model spread. The better-performing observational and model datasets are different in different parts of the CONUS.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

WearGP: A computationally efficient machine learning framework for local erosive wear predictions via nodal Gaussian processes

Computational fluid dynamics (CFD)-based wear predictions are computationally expensive to evaluate, even with a high-performance computing infrastructure. Thus, it is difficult to provide accurate local wear predictions in a timely manner. Data-driven approaches provide a more computationally efficient way to approximate the CFD wear predictions without running the actual CFD wear models. In this paper, a machine learning (ML) approach, termed WearGP, is presented to approximate the 3D local wear predictions, using numerical wear predictions from steady-state CFD simulations as training and testing datasets. The proposed framework is built on Gaussian process (GP) and utilized to predict wear in a much shorter time. The WearGP framework can be segmented into three stages. At the first stage, the training dataset is built by using a number of CFD simulations in the order of O (10 2 ). At the second stage, the data cleansing and data mining process are performed, where the nodal wear solutions are extracted from the solution database to build a training dataset. At the third stage, the wear predictions are made, using trained GP processes. Two CFD case studies including 3D slurry pump impeller and casing are used to demonstrate the WearGP framework, in which 144 training and 40 testing data points are used to train and test the proposed method, respectively. The numerical accuracy, computational effiency and effectiveness between the WearGP framework and CFD wear model for both slurry pump impellers and casings are compared. It is shown that the WearGP framework can achieve highly accurate results that are comparable with the CFD results, with a relatively small size training dataset, with a computational time reduction in an order of 10 5 to 10 6 .

42 ENGINEERING↗

Uniform-in-phase-space data selection with iterative normalizing flows

Improvements in computational and experimental capabilities are rapidly increasing the amount of scientific data that are routinely generated. In applications that are constrained by memory and computational intensity, excessively large datasets may hinder scientific discovery, making data reduction a critical component of data-driven methods. Datasets are growing in two directions: the number of data points and their dimensionality. Whereas dimension reduction typically aims at describing each data sample on lower-dimensional space, the focus here is on reducing the number of data points. A strategy is proposed to select data points such that they uniformly span the phase-space of the data. The algorithm proposed relies on estimating the probability map of the data and using it to construct an acceptance probability. An iterative method is used to accurately estimate the probability of the rare data points when only a small subset of the dataset is used to construct the probability map. Instead of binning the phase-space to estimate the probability map, its functional form is approximated with a normalizing flow. Therefore, the method naturally extends to high-dimensional datasets. The proposed framework is demonstrated as a viable pathway to enable data-efficient machine learning when abundant data are available.

97 MATHEMATICS AND COMPUTING↗

How does ion temperature gradient turbulence depend on magnetic geometry? Insights from data and machine learning

Magnetic geometry has a significant effect on the level of turbulent transport in fusion plasmas. Here, we model and analyse this dependence using multiple machine learning methods and a dataset of >200 000 nonlinear gyrokinetic simulations of ion-temperature-gradient turbulence in diverse non-axisymmetric geometries. The dataset is generated using a large collection of both optimised and randomly generated stellarator equilibria. At fixed gradients and other input parameters, the turbulent heat flux varies between geometries by several orders of magnitude. Trends are apparent among the configurations with particularly high or particularly low heat flux. Regression and classification techniques from machine learning are then applied to extract patterns in the dataset. Due to a symmetry of the gyrokinetic equation, the heat flux and regressions thereof should be invariant to translations of the raw features in the parallel coordinate, similar to translation invariance in computer vision applications. Multiple regression models including convolutional neural networks (CNNs) and decision trees can achieve reasonable predictive power for the heat flux in held-out test configurations, with highest accuracy for the CNNs. Using Spearman correlation, sequential feature selection and Shapley values to measure feature importance, it is consistently found that the most important geometric lever on the heat flux is the flux surface compression in regions of bad curvature. The second most important geometric feature relates to the magnitude of geodesic curvature. These two features align remarkably with surrogates that have been proposed based on theory, while the methods here allow a natural extension to more features for increased accuracy. The dataset, released with this publication, may also be used to test other proposed surrogates, and we find that many previously published proxies do correlate well with both the heat flux and stability boundary.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗