Email
中文
Home
About Us
About Us
Contact Us
Faculty
Faculty
Staff
TA&RA
Graduate Students
Visitors
Post Doctorals
Directories
Education
Undergraduate
Graduate
Introduction to basic mathematics course
First courses
Research
Research Areas
Conference
Colloquium
Math Department Invited Talks
Research Seminars
Student Seminars
Journals
Events
Current
Calendar
Past
News
News
Recruitment
Research & Teaching Positions
Postdocs
Staff
Download
Public
Private
Math Public
Campus Affairs
Join Us
E-hall
Announcement
News
Campus Map
Faculty & Staff
Library
Calendar
Admissions
Visit
Research
Research Areas
Conference
Colloquium
Math Department Invited Talks
Research Seminars
All
Algebra & Combinatorics Seminar
Computational & Applied Math Seminar
Dynamical Systems Seminar
Financial Math Seminar
Geometry & Topology Seminar
Graduate Student Colloquium
Number Theory Seminar
PDE Seminar
Probability & Statistics Seminar
Student Seminars
Journals
Computational & Applied Math Seminar
基于机器学习的非嵌入式降阶建模和气动力预测
Speaker: 段俊明(洛桑联邦理工学院)
Time: Jul 1, 2023, 15:00-16:00
Location: 理学院大楼M1001
摘要
:
降阶建模对于多查询或实时问题是不可或缺的研究手段,然而,为含时的参数化问题构建高效的降阶模型仍然面临许多挑战。本报告的第一部分将介绍一种针对含时参数化问题的非嵌入式多步降阶建模方法,它是数据驱动的,分为线下和线上两个部分。线下阶段采用机器学习技术进行有效的数据降维,将高维数据投影到低维隐空间/子流形上,并利用动态模分解方法对隐空间的动力学进行建模。线上阶段先在时间方向进行演化并在参数空间插值得到新的参数处的低维数据,然后重构出高维空间中的解。本报告的第二部分将讨论我们提出的系统性地构造非嵌入式实时气动力预测模型的方法。该方法基于数据融合,将来自数值模拟、实验和实时压力传感器输入的数据结合;先利用数值模拟数据找出最具代表性的传感器位置并建立线性预测模型,之后采用实验数据和机器学习技术进行校正,得到更为准确的非线性模型。该模型将实时压力传感器数据作为输入,可在微秒量级内给出准确的气动力预测,用于无人机导航。