课题基金 / 基金详情

GEM: Explorative Global-To Kinetic-Scale Modeling of Collisionless Shocks Using Physics-Informed Data Mining and Machine Learning

GEM: Explorative Global-To Kinetic-Scale Modeling of Collisionless Shocks Using Physics-Informed Data Mining and Machine Learning
GEM:使用物理信息数据挖掘和机器学习对无碰撞冲击进行探索性全局到动力学尺度建模
批准号:
2225463
负责人:
Drew Turner
金额:
$59.71万
依托单位:
依托单位国家:
美国
项目类别:
Standard Grant
财政年份:
2022
资助国家:
美国
项目状态:
未结题
起止时间:
2022-09-01 至 2025-08-31

项目摘要

项目成果

相似基金

相关文献

中文摘要
翻译
无碰撞冲击波发生在整个宇宙的空间等离子体中,被认为是加速宇宙射线的主要机制之一。了解在空间等离子体无碰撞冲击中起作用的物理学对许多基础科学领域具有广泛的影响,从太阳和空间物理学到行星科学和天体物理学。在地球上,太阳风等离子体在地球磁场周围偏转形成弓形激波,为探索这种无碰撞激波的本质提供了一个理想的自然实验室。这项工作有望使用最先进的机器学习和人工智能技术和工具,应用于10,000颗卫星穿越地球的弓形激波。目的是开发一系列数据驱动和物理信息的无碰撞冲击模型,不仅可以很好地捕捉地球自身弓形冲击的微观到宏观性质,而且还可以应用于更好地理解其他行星系统的冲击(包括其他恒星系统中的外行星),太阳和恒星冲击,以及其他更极端的天体物理冲击。如果成功,这些模型将证明具有变革性,使我们能够在我们无法立即访问的系统中产生和探索模拟冲击的例子,在太阳和磁层空间等离子体物理学以及行星和天体物理学之间建立新的跨学科联系。此外,这项工作涉及一个多元化的领导团队,并将为服务不足的学术界提供专门的研究机会。研究中心围绕着使用一对参数化的经验模型进行开发和探索:i)地球三维(3D),全球弓形激波,ii)离子动力学尺度下的一般无碰撞激波。模型将使用先进的数据挖掘和最先进的物理信息机器学习工具开发,这些工具应用于NASA的磁层多尺度(MMS)和太阳风数据集。MMS的大数据集是机器学习应用的理想选择,因为它伴随着“科学家在循环”(SITL)报告的数据集,它有效地作为专家确定且易于挖掘的验证数据集。一旦作为研究的一部分开发,模型i)有望成为第一个经验的(即,计算成本低廉),参数化的全球三维地球弓形激波模型,以准确捕获关键的准平行和准垂直激波状态及其各自的全球弓形激波表面畸变。模型ii)将通过将机器学习应用程序与空间等离子体中无碰撞冲击的基本物理原理相结合来开发,建立一个真正的物理信息机器学习模型,理想地提供准确的预测,不仅像那些用于训练模型的冲击(即,各种驾驶条件下的地球弓形冲击),而且还提供其他行星(和系外行星)弓形冲击的外推预测能力,无碰撞太阳冲击,行星际和日光层(和天体层)冲击,以及其他无碰撞的天体物理冲击。该奖项反映了美国国家科学基金会的法定使命,并通过使用基金会的知识价值和更广泛的影响审查标准进行评估,被认为值得支持。
英文摘要
Collisionless shock waves occur in space plasmas throughout the Universe and are one of the leading mechanisms considered responsible for accelerating cosmic rays. Understanding the physics at work in collisionless shocks in space plasmas has broad impacts on a number of fundamental scientific fields, from solar and space physics to planetary sciences and astrophysics. At Earth, a bow shock forms from the deflection of solar wind plasma around the planet's magnetic field, offering an ideal natural laboratory to explore the nature of such collisionless shocks. This effort promises to use state-of-the-art machine learning and artificial intelligence techniques and tools applied to 10,000 satellite crossings of Earth's bow shock. The intent is to develop a series of data-driven and physics-informed models of collisionless shocks that not only well capture the microscopic to macroscopic nature of Earth's own bow shock but may also be applied to a better understanding of shocks at other planetary systems (including exo-planets in other stellar systems), solar and stellar shocks, and other more extreme astrophysical shocks. If successful, these models will prove transformative by enabling us to produce and explore simulated examples of shocks in systems that we have no immediate access to, establishing a new interdisciplinary connection spanning between solar and magnetospheric space plasma physics and planetary and astrophysics. Furthermore, the effort involves a diverse leadership team and will provide dedicated research opportunities for underserved academic communities.The research centers around the development of and exploration using a pair of parameterized, empirical models of i) Earth's three-dimensional (3D), global bow shock, and ii) general collisionless shocks at ion-kinetic scales. Models will be developed using advanced data mining and state-of-the-art physics-informed machine learning tools applied to NASA's Magnetospheric Multiscale (MMS) and solar wind datasets. MMS' greater dataset is ideal for machine learning applications since it is accompanied by a dataset of "scientist-in-the-loop" (SITL) reports, which effectively serve as an expert-determined and easily minable validation dataset. Once developed as part of the research, model i) promises to be the first empirical (i.e., computationally inexpensive), parameterized, global-3D model of Earth's bow shock to accurately capture the critical quasi-parallel and quasi-perpendicular shock regimes and their respective distortion of the global bow shock surface. Model ii) will be developed by coupling machine learning applications to fundamental physical principles of collisionless shocks in space plasmas, establishing a genuine physics-informed machine learning model to ideally offer accurate predictions of not only shocks like those used to train the model (i.e., Earth's bow shock under a variety of driving conditions) but also extrapolative predictive capabilities for other planetary (and exoplanetary) bow shocks, collisionless solar shocks, interplanetary and heliospheric (and astrospheric) shocks, and other collisionless astrophysical shocks.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
期刊论文(0)
专著(0)
科研奖励(0)
会议论文
海外基金