SlideShare a Scribd company logo

Parameter estimation

M
Mohamed Mohamed El-Sayed

This document discusses various methods for parameter estimation of discrete linear systems, including deterministic and stochastic methods. Deterministic methods described include the simplified algorithm, Kaczmarz's algorithm, projection algorithm, batch least squares, weighted least squares, recursive least squares, and recursive least squares with exponential forgetting. Stochastic methods described include extended recursive least squares, extended recursive least squares with exponential forgetting, and modified extended recursive least squares. The goal of these methods is to estimate parameters of linear system transfer functions from input-output data in an online, recursive fashion.

Engineering
Related topics:
Adaptive Control Research
1 of 8
Downloaded 24 times
1
2
3
4
5
6
7
8
Contents 1.1 Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Deterministic Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.1 Simplified Algorithm (SA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.2 Kaczmarz’s Algorithm (KA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.3 Projection Algorithm (PA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.4 Batch Least Square (BLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.5 Weighted Least Square (WLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.6 Recursive Least Square (RLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.7 Recursive Least Square with Exponential Forgetting (RLS with λ) . . . . . . . . . . . . 4 1.2.8 Recursive Least Square with Varying Exponential Forgetting (RLS with varying λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3 Stochastic Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3.1 Extended Recursive Least Square (ERLS) . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3.2 Extended Recursive Least Square with Exponential Forgetting (ERLS with λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3.3 Extended Recursive Least Square with Varying Exponential Forgetting (ERLS with varying λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.3.4 Modified Extended Recursive Least Square (MERLS) . . . . . . . . . . . . . . . . . . . 6 1.3.5 Modified Extended Recursive Least Square with Exponential Forgetting (MERLS with λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.3.6 Modified Extended Recursive Least Square with Varying Exponential Forgetting (MERLS with varying λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.4 MATLAB Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.6 Contacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1
1.1. PARAMETER ESTIMATION Adaptive Control 1.1 Parameter Estimation On-line determination of process parameters is a key element in adaptive control. A recursive parameter estimator appears explicitly as a component of a self-tuning regulator. Parameter estimation also occurs implicitly in a model-reference adaptive controller. This section presents some methods for real-time parameter estimation. It is useful to view parameter estimation in the broader context of system identification. The key elements of system identification are selection of model structure, experiment design, parameter estimation, and validation. Since system identification is executed automatically in adaptive systems. it is essential to have a good understanding of all aspects of the problem. Selection of model structure and parameterization are fundamental issues. Simple transfer function models will be used in this chapter. The identification problems are simplified significantly if the models are linear in the parameters. This is about parameter estimation of discrete linear system transfer function using linear estimation methods. 1.2 Deterministic Parameter Estimation G(z−1 ) = z−dB(z−1) A(z−1) B(z−1 ) = b0 + b1z−1 + ... + bnbz−nb A(z−1 ) = 1 + a1z−1 + ... + anaz−na y(k) = φT (k)ˆθ(k − 1) φT (k) = [−y(k − 1) − y(k − 2) ... − y(k − na) u(k − d)u(k − d − 1) ... u(k − d − nb)] ˆθ(k − 1) = a1 a2 ... ana b0 b1 ... bnb T number of knowns = nu = na + nb + 1 n = Max(na, nb + d) number of sampled data = N > nu + n − 1 NOTE : Estimation stats from time n + 1 1.2.1 Simplified Algorithm (SA) φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] ˆθ(k + 1) = ˆθ(k) + φ(k + 1) φT (k + 1)φ(k + 1) y(k + 1) − φT (k + 1)ˆθ(k) 1.2.2 Kaczmarz’s Algorithm (KA) φT (k + 1) = [ −y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] ˆθ(k + 1) = ˆθ(k) + γφ(k + 1) φT (k + 1)φ(k + 1) y(k + 1) − φT (k + 1)ˆθ(k) α ≥ 0 0 < γ < 2 Mohamed Mohamed El-Sayed Atyya Page 2 of 8
1.2. DETERMINISTIC PARAMETER ESTIMATION Adaptive Control 1.2.3 Projection Algorithm (PA) φT (k + 1) = [ −y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] ˆθ(k + 1) = ˆθ(k) + γφ(k + 1) α + φT (k + 1)φ(k + 1) y(k + 1) − φT (k + 1)ˆθ(k) 1.2.4 Batch Least Square (BLS) ψ =     −y(k − 1) ... −y(k − na) u(k − d) ... u(k − d − nb) −y(k) ... −y(k − na + 1) u(k − d + 1) ... u(k − d − nb + 1) . ... . . ... . −y(N − 1) ... −y(N − na) u(N − d) ... u(N − d − nb)     Y = y(k) y(k + 1) ... y(N) T ∴ θ = ψT ψ −1 ψT Y 1.2.5 Weighted Least Square (WLS) θ = ψT Wψ −1 ψT WY Where W is a diagonal matrix with wii = γN−n−i−1 and γ < 1 1.2.6 Recursive Least Square (RLS) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] K(k + 1) = P(k)φ(k + 1) I + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 Mohamed Mohamed El-Sayed Atyya Page 3 of 8
1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control 1.2.7 Recursive Least Square with Exponential Forgetting (RLS with λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] K(k + 1) = P(k)φ(k + 1) λ + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k)/λ P(k + 1) = P(k + 1) + PT (k + 1) 2 1.2.8 Recursive Least Square with Varying Exponential Forgetting (RLS with varying λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu λ(k) = 0.3 → 0.999 φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] K(k + 1) = P(k)φ(k + 1) λ(k) + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = λ(k) − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 λ(k + 1) = 1 − 1 − φT (k + 1)K(k + 1) 2(k + 1) σ2( )µ( ) NOTE : Update λ after n times 1.3 Stochastic Parameter Estimation y(k) = z−dB(z−1) A(z−1) u + C(z−1) A(z−1) = φT (k)ˆθ(k − 1) B(z−1 ) = b0 + b1z−1 + ... + bnbz−nb A(z−1 ) = 1 + a1z−1 + ... + anaz−na C(z−1 ) = 1 + c1z−1 + ... + cncz−nc ˆθ(k − 1) = [a1 a2 ... ana b0 b1 ... bnb c1 c2 ... cnc]T number of knowns = nu = na + nb + nc + 1 n = Max(na, nb + nc + d) number of sampled data = N > nu + n − 1 NOTE : Estimation stats from time n + 1 Mohamed Mohamed El-Sayed Atyya Page 4 of 8
1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control Figure 1.1: Noise is applied on the output signal 1.3.1 Extended Recursive Least Square (ERLS) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) I + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 1.3.2 Extended Recursive Least Square with Exponential Forgetting (ERLS with λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k)/λ P(k + 1) = P(k + 1) + PT (k + 1) 2 Mohamed Mohamed El-Sayed Atyya Page 5 of 8
1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control 1.3.3 Extended Recursive Least Square with Varying Exponential Forgetting (ERLS with varying λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu λ(k) = 0.3 → 0.999 φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ(k) + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = λ(k) − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 λ(k + 1) = 1 − 1 − φT (k + 1)K(k + 1) 2(k + 1) σ2( )µ( ) NOTE : Update λ after n times 1.3.4 Modified Extended Recursive Least Square (MERLS) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) I + φT (k + 1)P(k)φ(k + 1) −1 − (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) − (k + 1) + (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 Mohamed Mohamed El-Sayed Atyya Page 6 of 8
1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control 1.3.5 Modified Extended Recursive Least Square with Exponential Forgetting (MERLS with λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ + φT (k + 1)P(k)φ(k + 1) −1 − (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) − (k + 1) + (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k)/λ P(k + 1) = P(k + 1) + PT (k + 1) 2 1.3.6 Modified Extended Recursive Least Square with Varying Exponential Forgetting (MERLS with varying λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu λ(k) = 0.3 → 0.999 φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ(k) + φT (k + 1)P(k)φ(k + 1) −1 − (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) − (k + 1) + (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k + 1) P(k + 1) = λ(k) − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 λ− (k + 1) = 1 − 1 − φT (k + 1)K(k + 1) −2 (k + 1) σ2( −)µ( −) λ+ (k + 1) = 1 − 1 − φT (k + 1)K(k + 1) +2 (k + 1) σ2( +)µ( +) σ− = σ2( −) σ2( −) + σ2( +) σ+ = σ2( +) σ2( −) + σ2( +) λ(k + 1) = σ− λ− (k + 1) + σ+ λ+ (k + 1) NOTE : Update λ after n times Mohamed Mohamed El-Sayed Atyya Page 7 of 8
1.4. MATLAB CODES Adaptive Control 1.4 MATLAB Codes 1.2.1 http://goo.gl/Vddvtt 1.2.2 http://goo.gl/UwWFTW 1.2.3 http://goo.gl/tV4Ni6 1.2.4 http://goo.gl/rY2n7I 1.2.6 http://goo.gl/e7J2kq 1.2.7 http://goo.gl/3q6Yc6 1.2.8 http://goo.gl/SCPvEW 1.3.1 http://goo.gl/JnrdNh 1.3.2 http://goo.gl/xjpHha 1.3.3 http://goo.gl/6wVeuW 1.3.4 http://goo.gl/vuKeaL 1.3.5 http://goo.gl/mL0RCz 1.3.6 http://goo.gl/vzViYE 1.5 References 1. Karl Johan Astrom, Adaptive Control, 2nd Edition. 2. David I. Wilson, Advanced Control using MATLAB or Stabilising the unstabilisable, Auckland University of Technology, New Zealand, May 15, 2015 1.6 Contacts mohamed.atyya94@eng-st.cu.edu.eg Mohamed Mohamed El-Sayed Atyya Page 8 of 8

More Related Content

PPTX
Introduction to Graph Theory
Premsankar Chakkingal
 
PDF
Introduction to optimization Problems
Electronics & Communication Staff SCU Suez Canal University
 
PDF
UMAP - Mathematics and implementational details
Umberto Lupo
 
PDF
Fuzzy sets and operators
Siksha 'O' Anusandhan (Deemed to be University )
 
PPTX
Fuzzy c means manual work
Dr.E.N.Sathishkumar
 
PPTX
Numerical Method 2
Rokonuzzaman Rony
 
PPT
Bayseian decision theory
sia16
 
PDF
Multiobjective optimization and trade offs using pareto optimality
Amogh Mundhekar
 
Introduction to Graph Theory
Premsankar Chakkingal
 
Introduction to optimization Problems
Electronics & Communication Staff SCU Suez Canal University
 
UMAP - Mathematics and implementational details
Umberto Lupo
 
Fuzzy sets and operators
Siksha 'O' Anusandhan (Deemed to be University )
 
Fuzzy c means manual work
Dr.E.N.Sathishkumar
 
Numerical Method 2
Rokonuzzaman Rony
 
Bayseian decision theory
sia16
 
Multiobjective optimization and trade offs using pareto optimality
Amogh Mundhekar
 

What's hot (20)

PDF
An Overview of Simple Linear Regression
Georgian Court University
 
PPTX
Gauss jordan method.pptx
RehmanRasheed3
 
PPTX
Basics of Integration and Derivatives
Faisal Waqar
 
PDF
Visual Explanation of Ridge Regression and LASSO
Kazuki Yoshida
 
PPTX
Simpson’s one third and weddle's rule
zahid6
 
PDF
Gradient Descent, Back Propagation, and Auto Differentiation - Advanced Spark...
Chris Fregly
 
PDF
08 Machine Learning Maximum Aposteriori
Andres Mendez-Vazquez
 
PDF
Application of linear transformation in computer
Favour Chukwuedo
 
PDF
Lesson 25: Unconstrained Optimization I
Matthew Leingang
 
PPT
System Of Linear Equations
saahil kshatriya
 
PDF
5. Linear Algebra for Machine Learning: Singular Value Decomposition and Prin...
Ceni Babaoglu, PhD
 
PDF
Gomory's cutting plane method
Rajesh Piryani
 
PDF
Fractional calculus and applications
PlusOrMinusZero
 
PPT
Introduction to Optimization.ppt
MonarjayMalbog1
 
PPT
1st order differential equations
Nisarg Amin
 
PDF
Special functions
Dr. Nirav Vyas
 
PPTX
inverse z transform
Zlatan Ahmadovic
 
PPTX
simpion's 3/8 rule
sharma230399
 
PPTX
Introduction to optimization
SambandamK1
 
PDF
Convex Optimization
adil raja
 
An Overview of Simple Linear Regression
Georgian Court University
 
Gauss jordan method.pptx
RehmanRasheed3
 
Basics of Integration and Derivatives
Faisal Waqar
 
Visual Explanation of Ridge Regression and LASSO
Kazuki Yoshida
 
Simpson’s one third and weddle's rule
zahid6
 
Gradient Descent, Back Propagation, and Auto Differentiation - Advanced Spark...
Chris Fregly
 
08 Machine Learning Maximum Aposteriori
Andres Mendez-Vazquez
 
Application of linear transformation in computer
Favour Chukwuedo
 
Lesson 25: Unconstrained Optimization I
Matthew Leingang
 
System Of Linear Equations
saahil kshatriya
 
5. Linear Algebra for Machine Learning: Singular Value Decomposition and Prin...
Ceni Babaoglu, PhD
 
Gomory's cutting plane method
Rajesh Piryani
 
Fractional calculus and applications
PlusOrMinusZero
 
Introduction to Optimization.ppt
MonarjayMalbog1
 
1st order differential equations
Nisarg Amin
 
Special functions
Dr. Nirav Vyas
 
inverse z transform
Zlatan Ahmadovic
 
simpion's 3/8 rule
sharma230399
 
Introduction to optimization
SambandamK1
 
Convex Optimization
adil raja
 
Ad

Similar to Parameter estimation (20)

PDF
Intro to Quant Trading Strategies (Lecture 6 of 10)
Adrian Aley
 
PDF
Forecasting With An Adaptive Control Algorithm
shwetakarsh
 
PDF
PONDICHERRY UNIVERSITY DEPARTMENT OF ELECTRONICS ENGINEERING.pdf
AWANISHKUMAR84
 
PPTX
lecture_09.pptx
PeruruFamidaNajumun
 
PDF
APPROACHES IN USING EXPECTATIONMAXIMIZATION ALGORITHM FOR MAXIMUM LIKELIHOOD ...
cscpconf
 
PPTX
CMS 2024 lecture 9 Kalman filters for identification.pptx
MasoudEsmaili5
 
PDF
Adaptive pi based on direct synthesis nishant
Nishant Parikh
 
PDF
Section9 stochastic
cairo university
 
PPTX
Adaptive filter
A. Shamel
 
PDF
Relatório
Manuel Beja da Costa
 
PDF
Discrete Kalman Filter (DKF)
Mohamed Mohamed El-Sayed
 
PDF
Lec_13.pdf
abbas miry
 
PDF
Development of Robust Adaptive Inverse models using Bacterial Foraging Optimi...
IDES Editor
 
PDF
Paper id 26201481
IJRAT
 
PDF
Fast parallelizable scenario-based stochastic optimization
Pantelis Sopasakis
 
PDF
tsoulkas_cumulants
Vasilis Tsoulkas
 
PDF
Line Search Techniques by Fibonacci Search
inventionjournals
 
PDF
Tail Probabilities for Randomized Program Runtimes via Martingales for Higher...
Satoshi Kura
 
PDF
Reinforcement Learning: Hidden Theory and New Super-Fast Algorithms
Sean Meyn
 
PDF
A walk through the intersection between machine learning and mechanistic mode...
JuanPabloCarbajal3
 
Intro to Quant Trading Strategies (Lecture 6 of 10)
Adrian Aley
 
Forecasting With An Adaptive Control Algorithm
shwetakarsh
 
PONDICHERRY UNIVERSITY DEPARTMENT OF ELECTRONICS ENGINEERING.pdf
AWANISHKUMAR84
 
lecture_09.pptx
PeruruFamidaNajumun
 
APPROACHES IN USING EXPECTATIONMAXIMIZATION ALGORITHM FOR MAXIMUM LIKELIHOOD ...
cscpconf
 
CMS 2024 lecture 9 Kalman filters for identification.pptx
MasoudEsmaili5
 
Adaptive pi based on direct synthesis nishant
Nishant Parikh
 
Section9 stochastic
cairo university
 
Adaptive filter
A. Shamel
 
Relatório
Manuel Beja da Costa
 
Discrete Kalman Filter (DKF)
Mohamed Mohamed El-Sayed
 
Lec_13.pdf
abbas miry
 
Development of Robust Adaptive Inverse models using Bacterial Foraging Optimi...
IDES Editor
 
Paper id 26201481
IJRAT
 
Fast parallelizable scenario-based stochastic optimization
Pantelis Sopasakis
 
tsoulkas_cumulants
Vasilis Tsoulkas
 
Line Search Techniques by Fibonacci Search
inventionjournals
 
Tail Probabilities for Randomized Program Runtimes via Martingales for Higher...
Satoshi Kura
 
Reinforcement Learning: Hidden Theory and New Super-Fast Algorithms
Sean Meyn
 
A walk through the intersection between machine learning and mechanistic mode...
JuanPabloCarbajal3
 
Ad

More from Mohamed Mohamed El-Sayed (20)

PDF
Optimal control systems
Mohamed Mohamed El-Sayed
 
PDF
Neural networks and deep learning
Mohamed Mohamed El-Sayed
 
PDF
Numerical solutions for ode (differential equations)
Mohamed Mohamed El-Sayed
 
PDF
Curve fitting
Mohamed Mohamed El-Sayed
 
PDF
Interpolation
Mohamed Mohamed El-Sayed
 
PDF
Numerical integration
Mohamed Mohamed El-Sayed
 
PDF
Numerical differentiation
Mohamed Mohamed El-Sayed
 
PDF
Numerical solutions for 1 non linear eq and system of non linear eqs
Mohamed Mohamed El-Sayed
 
PDF
Numerical solutions for linear system of equations
Mohamed Mohamed El-Sayed
 
PDF
Worm gears
Mohamed Mohamed El-Sayed
 
PDF
Bevel gears
Mohamed Mohamed El-Sayed
 
PDF
Helical gears
Mohamed Mohamed El-Sayed
 
PDF
Brakes
Mohamed Mohamed El-Sayed
 
PDF
Clutches
Mohamed Mohamed El-Sayed
 
PDF
Springs
Mohamed Mohamed El-Sayed
 
PDF
Chain drives
Mohamed Mohamed El-Sayed
 
PDF
V belt and rope drives
Mohamed Mohamed El-Sayed
 
PDF
Flat belt pulleys
Mohamed Mohamed El-Sayed
 
PDF
Flat belt drives
Mohamed Mohamed El-Sayed
 
PDF
Power screws
Mohamed Mohamed El-Sayed
 
Optimal control systems
Mohamed Mohamed El-Sayed
 
Neural networks and deep learning
Mohamed Mohamed El-Sayed
 
Numerical solutions for ode (differential equations)
Mohamed Mohamed El-Sayed
 
Curve fitting
Mohamed Mohamed El-Sayed
 
Interpolation
Mohamed Mohamed El-Sayed
 
Numerical integration
Mohamed Mohamed El-Sayed
 
Numerical differentiation
Mohamed Mohamed El-Sayed
 
Numerical solutions for 1 non linear eq and system of non linear eqs
Mohamed Mohamed El-Sayed
 
Numerical solutions for linear system of equations
Mohamed Mohamed El-Sayed
 
Worm gears
Mohamed Mohamed El-Sayed
 
Bevel gears
Mohamed Mohamed El-Sayed
 
Helical gears
Mohamed Mohamed El-Sayed
 
Brakes
Mohamed Mohamed El-Sayed
 
Clutches
Mohamed Mohamed El-Sayed
 
Springs
Mohamed Mohamed El-Sayed
 
Chain drives
Mohamed Mohamed El-Sayed
 
V belt and rope drives
Mohamed Mohamed El-Sayed
 
Flat belt pulleys
Mohamed Mohamed El-Sayed
 
Flat belt drives
Mohamed Mohamed El-Sayed
 
Power screws
Mohamed Mohamed El-Sayed
 

Recently uploaded (20)

PDF
Human-AI Collaboration: Balancing Agentic AI and Autonomy in Hybrid Systems
ijccsa
 
PDF
III.4.1.2_The_Space_Environment.p pdffdf
sittiporn2
 
PPTX
FINAL REVIEW FOR COPD DIANOSIS FOR PULMONARY DISEASE.pptx
rubeshbme
 
DOCX
573137875-Attendance-Management-System-original
AYUSHMANAV
 
PDF
737-MAX_SRG.pdf student reference guides
ssusere2119a1
 
PPTX
Engineering Ethics, Safety and Environment [Autosaved] (1).pptx
MehrabTaseenTalukder
 
PPT
Introduction, IoT Design Methodology, Case Study on IoT System for Weather Mo...
MariaRufina4
 
PDF
BMEC211 - INTRODUCTION TO MECHATRONICS-1.pdf
ryankakungu
 
PDF
A SYSTEMATIC REVIEW OF APPLICATIONS IN FRAUD DETECTION
IJNSA Journal
 
PPTX
UNIT-1 - COAL BASED THERMAL POWER PLANTS
Chandra Kumar S
 
PPTX
Fundamentals of safety and accident prevention -final (1).pptx
Palani kumar
 
PPTX
CARTOGRAPHY AND GEOINFORMATION VISUALIZATION chapter1 NPTE (2).pptx
abhi1361yadav
 
PDF
PREDICTION OF DIABETES FROM ELECTRONIC HEALTH RECORDS
ijaia
 
PDF
Well-logging-methods_new................
ZafriFarid
 
PPTX
Sustainable Sites - Green Building Construction
mhdumerali
 
PPT
Project quality management in manufacturing
BarMusaTetik
 
PPTX
UNIT 4 Total Quality Management .pptx
gokuld13012005
 
PDF
Embodied AI: Ushering in the Next Era of Intelligent Systems
Yu Huang
 
PPTX
Artificial Intelligence
dikshendrasarpate2
 
PPTX
M Tech Sem 1 Civil Engineering Environmental Sciences.pptx
ShriNRPrasad
 
Human-AI Collaboration: Balancing Agentic AI and Autonomy in Hybrid Systems
ijccsa
 
III.4.1.2_The_Space_Environment.p pdffdf
sittiporn2
 
FINAL REVIEW FOR COPD DIANOSIS FOR PULMONARY DISEASE.pptx
rubeshbme
 
573137875-Attendance-Management-System-original
AYUSHMANAV
 
737-MAX_SRG.pdf student reference guides
ssusere2119a1
 
Engineering Ethics, Safety and Environment [Autosaved] (1).pptx
MehrabTaseenTalukder
 
Introduction, IoT Design Methodology, Case Study on IoT System for Weather Mo...
MariaRufina4
 
BMEC211 - INTRODUCTION TO MECHATRONICS-1.pdf
ryankakungu
 
A SYSTEMATIC REVIEW OF APPLICATIONS IN FRAUD DETECTION
IJNSA Journal
 
UNIT-1 - COAL BASED THERMAL POWER PLANTS
Chandra Kumar S
 
Fundamentals of safety and accident prevention -final (1).pptx
Palani kumar
 
CARTOGRAPHY AND GEOINFORMATION VISUALIZATION chapter1 NPTE (2).pptx
abhi1361yadav
 
PREDICTION OF DIABETES FROM ELECTRONIC HEALTH RECORDS
ijaia
 
Well-logging-methods_new................
ZafriFarid
 
Sustainable Sites - Green Building Construction
mhdumerali
 
Project quality management in manufacturing
BarMusaTetik
 
UNIT 4 Total Quality Management .pptx
gokuld13012005
 
Embodied AI: Ushering in the Next Era of Intelligent Systems
Yu Huang
 
Artificial Intelligence
dikshendrasarpate2
 
M Tech Sem 1 Civil Engineering Environmental Sciences.pptx
ShriNRPrasad
 

Parameter estimation

  • 1. Contents 1.1 Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Deterministic Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.1 Simplified Algorithm (SA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.2 Kaczmarz’s Algorithm (KA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.3 Projection Algorithm (PA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.4 Batch Least Square (BLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.5 Weighted Least Square (WLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.6 Recursive Least Square (RLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.7 Recursive Least Square with Exponential Forgetting (RLS with λ) . . . . . . . . . . . . 4 1.2.8 Recursive Least Square with Varying Exponential Forgetting (RLS with varying λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3 Stochastic Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3.1 Extended Recursive Least Square (ERLS) . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3.2 Extended Recursive Least Square with Exponential Forgetting (ERLS with λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3.3 Extended Recursive Least Square with Varying Exponential Forgetting (ERLS with varying λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.3.4 Modified Extended Recursive Least Square (MERLS) . . . . . . . . . . . . . . . . . . . 6 1.3.5 Modified Extended Recursive Least Square with Exponential Forgetting (MERLS with λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.3.6 Modified Extended Recursive Least Square with Varying Exponential Forgetting (MERLS with varying λ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.4 MATLAB Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.6 Contacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1
  • 2. 1.1. PARAMETER ESTIMATION Adaptive Control 1.1 Parameter Estimation On-line determination of process parameters is a key element in adaptive control. A recursive parameter estimator appears explicitly as a component of a self-tuning regulator. Parameter estimation also occurs implicitly in a model-reference adaptive controller. This section presents some methods for real-time parameter estimation. It is useful to view parameter estimation in the broader context of system identification. The key elements of system identification are selection of model structure, experiment design, parameter estimation, and validation. Since system identification is executed automatically in adaptive systems. it is essential to have a good understanding of all aspects of the problem. Selection of model structure and parameterization are fundamental issues. Simple transfer function models will be used in this chapter. The identification problems are simplified significantly if the models are linear in the parameters. This is about parameter estimation of discrete linear system transfer function using linear estimation methods. 1.2 Deterministic Parameter Estimation G(z−1 ) = z−dB(z−1) A(z−1) B(z−1 ) = b0 + b1z−1 + ... + bnbz−nb A(z−1 ) = 1 + a1z−1 + ... + anaz−na y(k) = φT (k)ˆθ(k − 1) φT (k) = [−y(k − 1) − y(k − 2) ... − y(k − na) u(k − d)u(k − d − 1) ... u(k − d − nb)] ˆθ(k − 1) = a1 a2 ... ana b0 b1 ... bnb T number of knowns = nu = na + nb + 1 n = Max(na, nb + d) number of sampled data = N > nu + n − 1 NOTE : Estimation stats from time n + 1 1.2.1 Simplified Algorithm (SA) φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] ˆθ(k + 1) = ˆθ(k) + φ(k + 1) φT (k + 1)φ(k + 1) y(k + 1) − φT (k + 1)ˆθ(k) 1.2.2 Kaczmarz’s Algorithm (KA) φT (k + 1) = [ −y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] ˆθ(k + 1) = ˆθ(k) + γφ(k + 1) φT (k + 1)φ(k + 1) y(k + 1) − φT (k + 1)ˆθ(k) α ≥ 0 0 < γ < 2 Mohamed Mohamed El-Sayed Atyya Page 2 of 8
  • 3. 1.2. DETERMINISTIC PARAMETER ESTIMATION Adaptive Control 1.2.3 Projection Algorithm (PA) φT (k + 1) = [ −y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] ˆθ(k + 1) = ˆθ(k) + γφ(k + 1) α + φT (k + 1)φ(k + 1) y(k + 1) − φT (k + 1)ˆθ(k) 1.2.4 Batch Least Square (BLS) ψ =     −y(k − 1) ... −y(k − na) u(k − d) ... u(k − d − nb) −y(k) ... −y(k − na + 1) u(k − d + 1) ... u(k − d − nb + 1) . ... . . ... . −y(N − 1) ... −y(N − na) u(N − d) ... u(N − d − nb)     Y = y(k) y(k + 1) ... y(N) T ∴ θ = ψT ψ −1 ψT Y 1.2.5 Weighted Least Square (WLS) θ = ψT Wψ −1 ψT WY Where W is a diagonal matrix with wii = γN−n−i−1 and γ < 1 1.2.6 Recursive Least Square (RLS) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] K(k + 1) = P(k)φ(k + 1) I + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 Mohamed Mohamed El-Sayed Atyya Page 3 of 8
  • 4. 1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control 1.2.7 Recursive Least Square with Exponential Forgetting (RLS with λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] K(k + 1) = P(k)φ(k + 1) λ + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k)/λ P(k + 1) = P(k + 1) + PT (k + 1) 2 1.2.8 Recursive Least Square with Varying Exponential Forgetting (RLS with varying λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu λ(k) = 0.3 → 0.999 φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1)] K(k + 1) = P(k)φ(k + 1) λ(k) + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = λ(k) − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 λ(k + 1) = 1 − 1 − φT (k + 1)K(k + 1) 2(k + 1) σ2( )µ( ) NOTE : Update λ after n times 1.3 Stochastic Parameter Estimation y(k) = z−dB(z−1) A(z−1) u + C(z−1) A(z−1) = φT (k)ˆθ(k − 1) B(z−1 ) = b0 + b1z−1 + ... + bnbz−nb A(z−1 ) = 1 + a1z−1 + ... + anaz−na C(z−1 ) = 1 + c1z−1 + ... + cncz−nc ˆθ(k − 1) = [a1 a2 ... ana b0 b1 ... bnb c1 c2 ... cnc]T number of knowns = nu = na + nb + nc + 1 n = Max(na, nb + nc + d) number of sampled data = N > nu + n − 1 NOTE : Estimation stats from time n + 1 Mohamed Mohamed El-Sayed Atyya Page 4 of 8
  • 5. 1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control Figure 1.1: Noise is applied on the output signal 1.3.1 Extended Recursive Least Square (ERLS) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) I + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 1.3.2 Extended Recursive Least Square with Exponential Forgetting (ERLS with λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k)/λ P(k + 1) = P(k + 1) + PT (k + 1) 2 Mohamed Mohamed El-Sayed Atyya Page 5 of 8
  • 6. 1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control 1.3.3 Extended Recursive Least Square with Varying Exponential Forgetting (ERLS with varying λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu λ(k) = 0.3 → 0.999 φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ(k) + φT (k + 1)P(k)φ(k + 1) −1 (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) (k + 1) P(k + 1) = λ(k) − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 λ(k + 1) = 1 − 1 − φT (k + 1)K(k + 1) 2(k + 1) σ2( )µ( ) NOTE : Update λ after n times 1.3.4 Modified Extended Recursive Least Square (MERLS) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) I + φT (k + 1)P(k)φ(k + 1) −1 − (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) − (k + 1) + (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 Mohamed Mohamed El-Sayed Atyya Page 6 of 8
  • 7. 1.3. STOCHASTIC PARAMETER ESTIMATION Adaptive Control 1.3.5 Modified Extended Recursive Least Square with Exponential Forgetting (MERLS with λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ + φT (k + 1)P(k)φ(k + 1) −1 − (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) − (k + 1) + (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k + 1) P(k + 1) = I − K(k + 1)φT (k + 1) P(k)/λ P(k + 1) = P(k + 1) + PT (k + 1) 2 1.3.6 Modified Extended Recursive Least Square with Varying Exponential Forgetting (MERLS with varying λ) P0 = 106 I , or P0 = ψT ψ −1 , size(P) = nu x nu λ(k) = 0.3 → 0.999 φT (k + 1) = [−y(k) − y(k − 1) ... − y(k − na + 1) u(k − d + 1) u(k − d − 2) ... u(k − d − nb + 1) (k) (k − 1) ... (k − nc + 1)] K(k + 1) = P(k)φ(k + 1) λ(k) + φT (k + 1)P(k)φ(k + 1) −1 − (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k) ˆθ(k + 1) = ˆθ(k) + K(k + 1) − (k + 1) + (k + 1) = y(k + 1) − φT (k + 1)ˆθ(k + 1) P(k + 1) = λ(k) − K(k + 1)φT (k + 1) P(k) P(k + 1) = P(k + 1) + PT (k + 1) 2 λ− (k + 1) = 1 − 1 − φT (k + 1)K(k + 1) −2 (k + 1) σ2( −)µ( −) λ+ (k + 1) = 1 − 1 − φT (k + 1)K(k + 1) +2 (k + 1) σ2( +)µ( +) σ− = σ2( −) σ2( −) + σ2( +) σ+ = σ2( +) σ2( −) + σ2( +) λ(k + 1) = σ− λ− (k + 1) + σ+ λ+ (k + 1) NOTE : Update λ after n times Mohamed Mohamed El-Sayed Atyya Page 7 of 8
  • 8. 1.4. MATLAB CODES Adaptive Control 1.4 MATLAB Codes 1.2.1 http://goo.gl/Vddvtt 1.2.2 http://goo.gl/UwWFTW 1.2.3 http://goo.gl/tV4Ni6 1.2.4 http://goo.gl/rY2n7I 1.2.6 http://goo.gl/e7J2kq 1.2.7 http://goo.gl/3q6Yc6 1.2.8 http://goo.gl/SCPvEW 1.3.1 http://goo.gl/JnrdNh 1.3.2 http://goo.gl/xjpHha 1.3.3 http://goo.gl/6wVeuW 1.3.4 http://goo.gl/vuKeaL 1.3.5 http://goo.gl/mL0RCz 1.3.6 http://goo.gl/vzViYE 1.5 References 1. Karl Johan Astrom, Adaptive Control, 2nd Edition. 2. David I. Wilson, Advanced Control using MATLAB or Stabilising the unstabilisable, Auckland University of Technology, New Zealand, May 15, 2015 1.6 Contacts mohamed.atyya94@eng-st.cu.edu.eg Mohamed Mohamed El-Sayed Atyya Page 8 of 8