A Generalizable Physics-informed Learning Framework for Risk Probability Estimation

被引:0
|
作者
Wang, Zhuoyuan [1 ]
Nakahira, Yorie [1 ]
机构
[1] Carnegie Mellon Univ, Elect & Comp Engn Dept, Pittsburgh, PA 15213 USA
关键词
Stochastic safe control; physics-informed learning; risk probability estimation; SAFE EXPLORATION; REACHABILITY;
D O I
暂无
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Accurate estimates of long-term risk probabilities and their gradients are critical for many stochastic safe control methods. However, computing such risk probabilities in real-time and in unseen or changing environments is challenging. Monte Carlo (MC) methods cannot accurately evaluate the probabilities and their gradients as an infinitesimal devisor can amplify the sampling noise. In this paper, we develop an efficient method to evaluate the probabilities of long-term risk and their gradients. The proposed method exploits the fact that long-term risk probability satisfies certain partial differential equations (PDEs), which characterize the neighboring relations between the probabilities, to integrate MC methods and physics-informed neural networks. We provide theoretical guarantees of the estimation error given certain choices of training configurations. Numerical results show the proposed method has better sample efficiency, generalizes well to unseen regions, and can adapt to systems with changing parameters. The proposed method can also accurately estimate the gradients of risk probabilities, which enables first- and second-order techniques on risk probabilities to be used for learning and control.
引用
收藏
页数:13
相关论文
共 50 条
  • [41] A Southern Ocean supergyre as a unifying dynamical framework identified by physics-informed machine learning
    Sonnewald, Maike
    Reeve, Krissy Anne
    Lguensat, Redouane
    COMMUNICATIONS EARTH & ENVIRONMENT, 2023, 4 (01):
  • [42] Physics-informed deep learning framework to model intense precipitation events at super resolution
    B. Teufel
    F. Carmo
    L. Sushama
    L. Sun
    M. N. Khaliq
    S. Bélair
    A. Shamseldin
    D. Nagesh Kumar
    J. Vaze
    Geoscience Letters, 10
  • [43] A Grid-Induced and Physics-Informed Machine Learning CFD Framework for Turbulent Flows
    Chin Yik Lee
    Stewart Cant
    Flow, Turbulence and Combustion, 2024, 112 : 407 - 442
  • [44] A Southern Ocean supergyre as a unifying dynamical framework identified by physics-informed machine learning
    Maike Sonnewald
    Krissy Anne Reeve
    Redouane Lguensat
    Communications Earth & Environment, 4
  • [46] Physics-informed deep learning framework to model intense precipitation events at super resolution
    Teufel, B.
    Carmo, F.
    Sushama, L.
    Sun, L.
    Khaliq, M. N.
    Belair, S.
    Shamseldin, A.
    Kumar, D. Nagesh
    Vaze, J.
    GEOSCIENCE LETTERS, 2023, 10 (01)
  • [47] A framework based on physics-informed neural networks and extreme learning for the analysis of composite structures
    Yan, C. A.
    Vescovini, R.
    Dozio, L.
    COMPUTERS & STRUCTURES, 2022, 265
  • [48] A Grid-Induced and Physics-Informed Machine Learning CFD Framework for Turbulent Flows
    Lee, Chin Yik
    Cant, Stewart
    FLOW TURBULENCE AND COMBUSTION, 2024, 112 (02) : 407 - 442
  • [49] A physics-informed deep learning framework for spacecraft pursuit-evasion task assessment
    Yang, Fuyunxiang
    Yang, Leping
    Zhu, Yanwei
    CHINESE JOURNAL OF AERONAUTICS, 2024, 37 (05) : 363 - 376
  • [50] A physics-informed deep learning framework for spacecraft pursuit-evasion task assessment
    Fuyunxiang YANG
    Leping YANG
    Yanwei ZHU
    Chinese Journal of Aeronautics, 2024, 37 (05) : 363 - 376