Professional Documents
Culture Documents
ELEN 764 Power System Planning: Experimental Exploration Two
ELEN 764 Power System Planning: Experimental Exploration Two
Exploration Two
Charles Winley Jr.
Department of Electrical and Computer Engineering
North Carolina Agricultural and Technical State University
Greensboro, North Carolina, USA
cwinley@ncat.edu
1
3 Linear Least Squares 4 Weight Estimates
Linear least squares, first described by Carl
Friedrich Gauss around 1794 is a method used for
fitting a linear model to data. Least squares is of-
ten applied in statistical contexts, particularly re-
gression analysis. Linear least squares regression For exercise 2 the normal equations for Base,
is by far the most widely used modeling method, Growth, and the fourier coefficients for N =
due to its simplicity, the goals of linear least squares 1, 2, 3, and4 will be utilized to help develop a data
are to extract predictions from the measurements set . Using the calculated data set (X) and the orig-
and to reduce the effect of measurement errors. A inal PSL (y) you can now calculate the weight es-
fundamental assumption of the least squares is the timates using the OLAM Pseudo inverse. (equation
best-fit curve through the experimental data is one 8)
that exhibits the least square error from the given
set of data. In terms involving a single variable,
(x1 , y1 ), ...(xn , yn ) where the x’s are the indepen-
dent variables and y is the dependent variable we at-
tempt to use a fitting cure f(x). We will be using the
linear least squares equation (equation 1) to derive
model consisting of base and growth.
M
X b = (X T X)−1 X T ∗ y
w (8)
2
SSD(Base, Growth) = (P SLw −Base−Growth∗w)
w=1
(1)
M
∂SSD X
= −2 wi ∗ (P SLi − B − G ∗ wi )2 = 0
∂G i=1 SL = X ∗ w
Pd b (9)
(2)
M
∂SSD X
= −2 (P SLi − B − G ∗ wi )2 = 0 (3)
∂B i=1
X X X
G wi2 + B wi = wi ∗ P SLi (4)
X X In this exercise it was observed that as you in-
G wi + n ∗ B = P SLi (5)
creased the number of harmonics (N) from 1 har-
monic to 4 harmonics the Pd SL began to look more
P P P like the original PSL model. By calculating the per-
wi ∗ P SLi − wi P SLi
n
Growth = = 15.25 cent error each time the number of harmonics (N)
n wi2 − ( wi )2
P P
was increased I was able observe that the best re-
(6)
1 X X sults were present when either two or three harmon-
Base = ( P SLi − Growth wi ) = 7540 ics were used when calculating Pd SL. This variance
n
(7) I feel was due to the randomization of the data set.
Figure 1. PSLtrn and PSLtrn hat w/ N=1
Figure 2. PSLtrn and PSLtrn hat w/ N=2 Figure 4. PSLtrn and PSLtrn hat w/ N=4