100% found this document useful (1 vote)
481 views

Lecture 6 (Curve Fitting)

- General least squares (GLS) is an iterative process that can handle correlated observations with unequal precision and combined observations and parameters in condition equations. - GLS forms condition equations relating observations and unknown parameters and solves the equations using a matrix form (Av + BΔ = f) to estimate the parameters. - The concept of weighting accounts for the differing precision of observations by giving more weight to observations with higher precision. Weights are incorporated into the coefficient matrix and solution. - An example shows using GLS to estimate the parameters a and b in a straight line model y=ax+b by linearizing the model equation and solving the resulting matrices. Initial parameter estimates come from a line through two points.

Uploaded by

Kismet
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
100% found this document useful (1 vote)
481 views

Lecture 6 (Curve Fitting)

- General least squares (GLS) is an iterative process that can handle correlated observations with unequal precision and combined observations and parameters in condition equations. - GLS forms condition equations relating observations and unknown parameters and solves the equations using a matrix form (Av + BΔ = f) to estimate the parameters. - The concept of weighting accounts for the differing precision of observations by giving more weight to observations with higher precision. Weights are incorporated into the coefficient matrix and solution. - An example shows using GLS to estimate the parameters a and b in a straight line model y=ax+b by linearizing the model equation and solving the resulting matrices. Initial parameter estimates come from a line through two points.

Uploaded by

Kismet
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

General Least Squares

• Can handle combined


observations and parameters in
General Least Squares Adjustment the condition equations
(Curve Fitting) • No restriction on the number of
observations per equation

General Least Squares General Least Squares


• May be used for correlated • In the form:
observations with unequal
precision
• An iterative process Av+B∆= f
• Best used in solving curve-fitting
problems and coordinate-
transformation problems
Av = f v+B∆= f
Adjustment of Adjustment of Indirect
Observations Only Observations

Derivation of the GLS Derivation of the GLS


• Given a mathematical model, the • If we want to include unknown
minimum number of observations for parameters into the adjustment, we
its unique determination is n0 have to write an additional condition
• When n measurements for the model equation for each parameter
are acquired, such that n>n0, the • If u parameters are to be included,
redundancy is: the number of condition equations
r = n – n0 will be:
 for n observational variables, c=r+u
there exist r independent condition
equations

1
Derivation of the GLS Derivation of the GLS
• Lower limit: • Therefore,
-- u is equal to zero (u = 0)
-- number of condition equations is
equal to redundancy (c = r)
Lower Limit
0 ≤ u ≤ n0 Upper Limit

• Upper limit:
-- u is equal to n0 (u = n0)
-- number of condition equations is:
Adjustment of
Observations
r≤c≤n Adjustment of
Indirect
c = r + n0 = n Only Observations
 maximum number of
conditions is equal to the number of
observations

Derivation of the GLS The Concept of Weight


• In many problems, however, the 1 The smaller the variance,
precision α
σ2 the higher the precision
number of parameters in the
adjustment is neither zero nor n0
 the two special cases will not 1
weight α
be directly suitable σ2
 we will need to use the
GENERAL LEAST SQUARE Where:
TECHNIQUE σ 02
weight = W = σ02 = “reference variance”
σ2 = value that will make
the largest weight equal to 1

The Concept of Weight The Concept of Weight


 w1 0 0  σ 02 / σ 12 0 0 
  W = σ 02Σ −1
W =0 w2 
0 = 0 σ0 / σ2
2 2
0 
 
 0 0 wn   0 0 σ 02 / σ n2 
σ 12 0 0
 
Σ = covariance matrix =  0 σ 2 0 
2

1 / σ 12 0 0  0
   0 σ n2 
W =σ  0
2
0 1 / σ 22 0 
 0 0 1 / σ n 
2

1
Q= Σ Cofactor matrix (Q)
σ 02

2
Example Example
• For any of the points, the general
A straight line, y = ax + b, must be fitted
through three points. The following data are
equation of the straight line is:
given: F = ( y + v y ) − a ( x + vx ) − b = 0
Point X (cm) Y (cm) σx2 (cm2) σy2 (cm2)
1 2.00 3.20 0.04 0.10  non-linear in the unknowns!
2 4.00 4.00 0.04 0.08  least squares adjustment
3 6.00 5.00 0.04 0.08 cannot be used directly
Find the least square estimates of the  equation needs to be
parameters a and b, using GLS. linearized first…

Example Example
• The linearized form of the condition • In matrix form:
equation is:

∂F ∂F ∂F ∂F  vx  δ a 
( y − a0 x − b0 ) + vx + vy + δa+ δb = 0 [ −a0 1]   + [ − x −1]   = (− y + a0 x + b0 )
∂x ∂y ∂a ∂b v
 y δ b 

(−a0 )vx + (1)v y + (− x)δ a + (−1)δ b = (− y + a0 x + b0 )

Example Example
• The linearized equations (matrix • The initial approximations for a0 and
form) for all three points can be b0 can be obtained by using the
written as: equation of the straight line that
 vx1 
passes through any two of the three
  points.
 v y1 
 − a0 1 0 0 0 0    − x1 −1  ( − y1 + a0 x1 + b0 )  • Using points 2 and 3 for instance,
 0 vx2 δ a 
 0 −a0 1 0 0   +  − x2 −1   =  (− y2 + a0 x2 + b0 ) 
vy  δb
 0 0 0 0 − a0 1   2   − x3 −1    (− y3 + a0 x3 + b0 )  y2 = a0 x2 + b0
 vx3 
 
 v y3  y3 = a0 x3 + b0
 This is in the form: Av + B∆ = f

3
Example Example
• Solving for a0 and b0, • The initial matrices are therefore,
 −0.5 1 0 0 0 0
y − y2 5.00 − 4.00
a0 = 3 = = 0.500 A =  0 0 −0.5 1 0 0 
x3 − x2 6.00 − 4.00
 0 0 0 0 −0.5 1 
b0 = y2 − a0 x2 = 4.00 − 0.500(4.00) = 2.00
 −2 −1  −3.20 + 0.5(2.00) + 2.00   −0.20 
B =  −4 −1 f =  −4.00 + 0.5(4.00) + 2.00  =  0 
 −6 −1  −5.00 + 0.5(6.00) + 2.00   0 

Example Example
• The covariance matrix is: • From the general law of propagation
σ x21 0 0 0 0 0  of variances and covariances, the
 
0 σ y2 1
0 0 0 0  covariance matrix is transformed to:
 
0 0 σ x2 0 0 0 
Σ= 2

0

0 0 σ y2 2
0 0 

 0.11 0 0 
0 0 0 0 σ x2 0  
Σ e = AΣA =  0
t
0.09 0  cm 2
 
3

 0 0 0 0 0 σ y23 
0.04 0 0 0 
0 0
 0 0 0.09 
 0 0.10 0 0 
0 0

 0 0 0.04 0 0 0  2
Σ=  cm
• If the reference variance is σ 0 = 0.09
2
 0 0 0 0.08 0 0 
 0 0 0 0 0.04 0 
 
 0 0 0 0 0 0.08 

Example Example
• The equivalent weight matrix is To solve for the unknowns, we use the
therefore, formula:

We = σ 02Σ e−1 ( B W B) ∆ = B W f
T
e
T
e

 1 
 0.11 0 0 
  0.818 0 0 
We = (0.09)  0

1
0.09
0  =  0

1 0  N∆ = t
   0 0 1 
 0 1  
0
 0.09 
TRAINING CENTER FOR APPLIED GEODESY AND PHOTOGRAMMETRY
COLLEGE OF ENGINEERING, UNIVERSITY OF THE PHILIPPINES, DILIMAN

4
Example Example
• The solved matrices are therefore, • The corrected parameter values are:

55.272 11.636 
N =
11.636 2.818 
 a1 = a0 + δ a = 0.5 − 0.0482
0.3272 
t= 
b1 = b0 + δ b = 2.0 + 0.2571
0.1636 
 0.1384 −0.5715 a1 = 0.4518
N −1 =  
 −0.5715 2.7147  b1 = 2.2571
 −0.0482 
∆ = N −1t =    these are only the results from the 1st
 0.2571  iteration.

Example Example
After second iteration: To calculate the adjusted coordinates, we
use the derived equation for the residuals v:
δa= 3.58242E-05
δb= -0.000191062
v = QAtWe ( f − B∆ ) *
a= 0.451822
b= 2.256952
The equation of the line therefore is:
* For reference, you may read the derivation in the book
“Analysis and Adjustment of Survey Measurements”,
y = 0.4518x+2.2700 Mikhail and Gracie. Von Nostrand Reinhold: 1981.
pp. 242-246

Example Example
Adjusted coordinates of the three points:
 0.01  2.00  0.01   2.01
− 0.04 3.20 − 0.04 3.16 
       
POINT x(cm) y(cm)
 − 0.01 ^ 4.00  − 0.01 3.99  1 2.01 3.16
v=  l =l+v= + =   2 3.99 4.06
 0.06  4.00  0.06   4.06
 0.01  6.00  0.01   6.01 3 6.01 4.97
       
5.00  − 0.03 4.97 
− 0.03
Verify if these adjusted positions lie on
the line y = 0.4518x+2.2570

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy