SINGULAR VALUE DECOMPOSITION AND LEAST SQUARES ORBIT DETERMINATION

Similar documents
Development of an algorithm for the problem of the least-squares method: Preliminary Numerical Experience

ORBIT DETERMINATION OF LEO SATELLITES FOR A SINGLE PASS THROUGH A RADAR: COMPARISON OF METHODS

Contribution of ISON and KIAM space debris. space

Performance of a Dynamic Algorithm For Processing Uncorrelated Tracks

14 Singular Value Decomposition

LINEARIZED ORBIT COVARIANCE GENERATION AND PROPAGATION ANALYSIS VIA SIMPLE MONTE CARLO SIMULATIONS

Review of Some Concepts from Linear Algebra: Part 2

Numerical Optimization

Vector and Matrix Norms. Vector and Matrix Norms

Constrained optimization. Unconstrained optimization. One-dimensional. Multi-dimensional. Newton with equality constraints. Active-set method.

Optimization. Escuela de Ingeniería Informática de Oviedo. (Dpto. de Matemáticas-UniOvi) Numerical Computation Optimization 1 / 30

Astrodynamics 103 Part 1: Gauss/Laplace+ Algorithm

Constrained optimization: direct methods (cont.)

10-725/36-725: Convex Optimization Prerequisite Topics

A New Trust Region Algorithm Using Radial Basis Function Models

x n+1 = x n f(x n) f (x n ), n 0.

Solving Symmetric Indefinite Systems with Symmetric Positive Definite Preconditioners

Numerical Optimal Control Overview. Moritz Diehl

Discrete Ill Posed and Rank Deficient Problems. Alistair Boyle, Feb 2009, SYS5906: Directed Studies Inverse Problems 1

on space debris objects obtained by the

Improving the Convergence of Back-Propogation Learning with Second Order Methods

B5.6 Nonlinear Systems

Collision Prediction for LEO Satellites. Analysis of Characteristics

Chapter 3. Algorithm for Lambert's Problem

1 h 9 e $ s i n t h e o r y, a p p l i c a t i a n

Assignment #10: Diagonalization of Symmetric Matrices, Quadratic Forms, Optimization, Singular Value Decomposition. Name:

7 Principal Component Analysis

STATISTICAL ORBIT DETERMINATION

Scientific Computing: Optimization

linearly indepedent eigenvectors as the multiplicity of the root, but in general there may be no more than one. For further discussion, assume matrice

11 a 12 a 21 a 11 a 22 a 12 a 21. (C.11) A = The determinant of a product of two matrices is given by AB = A B 1 1 = (C.13) and similarly.

Linear Algebra Review

EXPANDING KNOWLEDGE ON REAL SITUATION AT HIGH NEAR-EARTH ORBITS

Radar-Optical Observation Mix

Long-Term Evolution of High Earth Orbits: Effects of Direct Solar Radiation Pressure and Comparison of Trajectory Propagators

Quadratic Programming

Notes on PCG for Sparse Linear Systems

AMS 529: Finite Element Methods: Fundamentals, Applications, and New Trends

ON SELECTING THE CORRECT ROOT OF ANGLES-ONLY INITIAL ORBIT DETERMINATION EQUATIONS OF LAGRANGE, LAPLACE, AND GAUSS

Linear Algebra and Eigenproblems

Space Surveillance with Star Trackers. Part II: Orbit Estimation

Outline. Relaxation. Outline DMP204 SCHEDULING, TIMETABLING AND ROUTING. 1. Lagrangian Relaxation. Lecture 12 Single Machine Models, Column Generation

Half of Final Exam Name: Practice Problems October 28, 2014

5 Linear Algebra and Inverse Problem

Linear Algebra Massoud Malek

Examples include: (a) the Lorenz system for climate and weather modeling (b) the Hodgkin-Huxley system for neuron modeling

AS3010: Introduction to Space Technology

Least Squares. Tom Lyche. October 26, Centre of Mathematics for Applications, Department of Informatics, University of Oslo

Orbit Determination Using Satellite-to-Satellite Tracking Data

Multigrid absolute value preconditioning

Linear Regression and Its Applications

Satellite Orbital Maneuvers and Transfers. Dr Ugur GUVEN

Optimization. Benjamin Recht University of California, Berkeley Stephen Wright University of Wisconsin-Madison

Linear Algebra. Session 12

The Big Picture. Discuss Examples of unpredictability. Odds, Stanisław Lem, The New Yorker (1974) Chaos, Scientific American (1986)

Algorithms for constrained local optimization

Part 3: Trust-region methods for unconstrained optimization. Nick Gould (RAL)

15 Singular Value Decomposition

Sparse Approximation of Signals with Highly Coherent Dictionaries

Numerical Methods I Solving Nonlinear Equations

Lecture 15 Perron-Frobenius Theory

Numerical tensor methods and their applications

RESEARCH ARTICLE. A strategy of finding an initial active set for inequality constrained quadratic programming problems

NORMS ON SPACE OF MATRICES

DEFINITION OF A REFERENCE ORBIT FOR THE SKYBRIDGE CONSTELLATION SATELLITES

Lecture 8. Principal Component Analysis. Luigi Freda. ALCOR Lab DIAG University of Rome La Sapienza. December 13, 2016

Numerical approximation for optimal control problems via MPC and HJB. Giulia Fabrini

Agenda. Fast proximal gradient methods. 1 Accelerated first-order methods. 2 Auxiliary sequences. 3 Convergence analysis. 4 Numerical examples

Elementary linear algebra

Examination paper for TMA4145 Linear Methods

Data dependent operators for the spatial-spectral fusion problem

8 th US/Russian Space Surveillance Workshop

CS6964: Notes On Linear Systems

MA2AA1 (ODE s): The inverse and implicit function theorem

Chapter 6 Nonlinear Systems and Phenomena. Friday, November 2, 12

Conditional Gradient (Frank-Wolfe) Method

The Launch of Gorizont 45 on the First Proton K /Breeze M

Mathematical optimization

Iterative Methods. Splitting Methods

Conjugate Gradient algorithm. Storage: fixed, independent of number of steps.

Applied Machine Learning for Biomedical Engineering. Enrico Grisan

MAT 419 Lecture Notes Transcribed by Eowyn Cenek 6/1/2012

CHAPTER 2: QUADRATIC PROGRAMMING

On Sun-Synchronous Orbits and Associated Constellations

Computational Methods. Eigenvalues and Singular Values

STA141C: Big Data & High Performance Statistical Computing

Invariant Manifolds of Dynamical Systems and an application to Space Exploration

Least Squares. Chapter Least Squares The Definition of Ordinary Least Squares

Numerical Methods I Non-Square and Sparse Linear Systems

Beyond Heuristics: Applying Alternating Direction Method of Multipliers in Nonconvex Territory

When Does the Uncertainty Become Non-Gaussian. Kyle T. Alfriend 1 Texas A&M University Inkwan Park 2 Texas A&M University

Parallel Algorithm for Track Initiation for Optical Space Surveillance

List of Tables. Table 3.1 Determination efficiency for circular orbits - Sample problem 1 41

Orbits in Geographic Context. Instantaneous Time Solutions Orbit Fixing in Geographic Frame Classical Orbital Elements

Coordinate Update Algorithm Short Course Proximal Operators and Algorithms

Introduction to Scientific Computing

Robust Principal Component Pursuit via Alternating Minimization Scheme on Matrix Manifolds

Preliminary Examination in Numerical Analysis

Numerical Methods I Eigenvalue Problems

17 Solution of Nonlinear Systems

Transcription:

SINGULAR VALUE DECOMPOSITION AND LEAST SQUARES ORBIT DETERMINATION Zakhary N. Khutorovsky & Vladimir Boikov Vympel Corp. Kyle T. Alfriend Texas A&M University

Outline Background Batch Least Squares Nonlinear effects Singular Value Decomposition Modified least squares with nonlinear effects Algorithm Results Conclusions

Background Batch least squares (BLS) is the standard orbit determination method in the US and Russia for catalog maintenance. BLS performance is generally satisfactory for catalog maintenance because it is a relatively linear problem. BLS has convergence problems when the nonlinear effects become significant, which are: Poor initial or a priori state. Long time between tracks, sparse measurements. Poor or abnormal measurements Difficulty of obtaining a good Hessian approximation in regions that are remote from the solution. Problem occurs mostly with IOD and UCT correlation Propose using SVD to resolve these problems by choosing the dimension of the minimization subspace at each step.

Least Squares Least squares is concerned with the minimization of Fx 1 2 m k i i1 j1 ŝ ij s ij ij x 2 gradf x gx 0 Define F ŝ ij ŝij ij, s ij Standard BLS reduces to x s ij x ij s s 11,..., s 1k1, s 21,..., s 2k2,..., s m1,..., s mkm 1 1 2 2 T ˆ ˆ T x s s x s s x b b, b sˆ s x T T T T b b s s g x F x b,, A x x x x x x A T Ax GN A T b T T

Revisit Least Squares Development Expand b in a Taylor series about a reference trajectory. b bx b x 1 2 b x xx 2 xt x.. x 2 xx b x b 2 b x A 2 b x x xx x 2 x 2 xx xx The least squares process reduces to A T A Bx A T b M B 2 M b x i b 2 i 2 s x i b x 2 i x x i1 i1 We propose an optimal strategy, which will initially move along the directions for which B is not significant, until it reaches the area of small residuals, where B can then be disregarded.

Singular Value Decomposition A USV T U T U V T V I S diag s 1, s 2,..., s n, s 1 s 2... s n T T g U b, y V x Sy g y g / s j j j Consider the probe vector y x k y, y,..., y,0,...,0 1 2 k k j Vy j1 Square of the normalized residual norm is m 2 k 2 k b Ax g j jk1 k k y v j T Need to find an index k such that the norm of the probe vector and the norm of the residual for this probe solution are small enough.

Proposed Procedure 1. Develop the matrix of trial vectors x k k j1 V j g j w j 2. For each trial vector compute the expected decrease of the least squares error function n k 2 g j 2 jk1 3. Check acceptability of each of the trial vectors. If not satisfied by all vectors normalize x j k by d min k x j c j If not satisfied, d j f c j k x,d min min j j f d j 4. Check relative decrease of the SVD method as we go to the next trial vector. If the inequality is satisfied then the trial vector x (k) is taken as the next iteration. 2 k1 2 k1 k 2 C 5. If the inequality in Step 4 is not satisfied then the previous trial vector is used. 6. After computing the least squares function with x (k) determine if the SVD method is converging sufficiently. Determine if F k1 F k If this inequality is satisfied go to the next step. If it is not satisfied then the modified least squares method is used because the Hessian is degenerate and the residuals need to be considered. F k1 C F

Residual Comparison

Residual Comparison

Residual Comparison

Results Geosynchronous satellite Angles only obs Short arc, 25 hour max ICs obtained using Laplace s method Theory is the Russian semi-analytic theory Table 1 O bservation Data Example 1 2 3 4 # of obs 7 5 8 4 Ob Time span (hr:min) 25:22 5:06 25:39 4:21 Table 2 Initial Conditions For the Orbit Determination Example Period Inclination Arg. of Per. Ascend Node Eccentricity (min) (deg) (deg) (deg) 1 1442.78 15.22 117.0-5.27 0.0046 2 1782.05 9.33-3.18 48.14 0.1688 3 1436.01 15.21-122.24-4.883 0.1029 4 1459.05 11.33 24.45 35.52 0.0090

Example 1 Fast Convergence Iter SS 1 5 2 6 3 6 4 6 5 6 6 6

Example 2 Intermediate Convergence Iter SS 1 1 2 4 3 5 4 5 5 5 6 5 7 5 8 6 9 6 10 6 11 6

Example 3 Slow Convergence Iter SS 1 1 2 1 3 3 4 5 5 5 6 5 7 5 8 5 9 5 10 5 11 5 12 6 13 6 14 6 15 6

Example 4 Slow Convergence Iter SS 1 1 2 4 3 4 4 5 5 4 6 5 7 5 8 6 9 6 10 6 11 6 12 6

Conclusions New method of orbit determination using singular value decomposition with least squares developed. A strategy using SVD and modified least squares that incorporates some nonlinear effects presented. Modified method used when Hessian is degenerate. SVD method applied to short arc, angles only 24-hour satellites shows significant improvement over standard least squares. As we look to the future of a catalog of more than 100,000 objects the US should look at Russian approaches to determine if they would help improve our catalog development and maintenance.