Beruflich Dokumente
Kultur Dokumente
I.
INTRODUCTION
208
17th IEEE Mediterranean Electrotechnical Conference, Beirut, Lebanon, 13-16 April 2014.
II.
text
text
P2J2
......
text
P1J2
text
text
...
PIJ1
text
PI22
text
text
PIJ2
PI2K
P2JK
...
PI21
P1JK
PIJK
(3)
2 (4)
A. Nonparametric Test
The purpose of probabilistic tests is to compare the effect of
sampling time on load variation patterns in terms of data
distribution. Nonparametric tests are used for comparison
purposes because different data samples may follow different
kind of probability distributions.
1) Two-sample Kolmogorov-Smirnov (KS) test
This is a non-parametric test and an empirical distribution
function is used to compare the data distribution of two
samples. The null hypothesis (0 ) for this test is defined
assuming that both samples belong to the same data
distribution. The result of this test is 1 if 0 is rejected at given
significance level , otherwise it is 0. With given i and j,
compared all observations k have been compared with each
other to verify the above test results. If x, y = 1,2, K, then by
using (5) this test can be formulated mathematically as shown
in (7) - (9).
(5)
1J
IJ
text
...
...
12
I2
P2J1
(2)
11
=
I1
PI1K
ij = 2
P221
...
...
= [ , 2 , ]
P211
PI12
If the load profile starts from mid night then will exactly
equal to and (2) can be rewritten as:
2 2
P1J1
PI11
:
time instant for the first sample
:
total number of samples
:
load variation profile for a typical day
...
...
= [ , + , +
2 + ( 1) ]
1 2
P121
(1)
() = () ( )
ij = 2
P111
(6)
209
17th IEEE Mediterranean Electrotechnical Conference, Beirut, Lebanon, 13-16 April 2014.
, > ()
=
=1 =1(, )
(7)
150
1,
=
0,
(8)
(9)
=
=1 =1(, )
=1
1
=
=1( )
6500
6000
50
5500
5000
10 15 20 25 30 35 40 45 50 55 60
Sampling Interval (Minutes)
4500
10000
9500
9000
8500
8000
75
7500
7000
50
6500
10 15 20 25 30 35 40 45 50 55 60
Sampling Interval (Minutes)
6000
By using (13) and (14), Eq. (3) and Eq. (6) can be
and as shown in (15) to (18).
represented in terms of
Here the data is reduced to 2 dimensions as the 3rd dimension
is eliminated by averaging the data using observations.
(11)
= [
,
2 ,
]
(12)
11
12
= [ , 2 , ]
11
=
1
7000
1
=
7500
75
8000
20
8500
150
1,
0,
9000
= =1 , = =1 (10)
, =
9500
20
10000
, = ( ) ( )
12
(15)
(16)
(17)
(18)
Here,
and are the matrices of organized
and
data sets.
The standard deviation provides the information about
the possible deviation or error with respect to the estimated
. A statistical measure for this deviation is
mean
introduced in this paper to see the effect of sampling time and
aggregation level. This statistical measure (ARSD) is
calculated using (19) and is organized in (20) for all
combinations of and .
(13)
(14)
210
17th IEEE Mediterranean Electrotechnical Conference, Beirut, Lebanon, 13-16 April 2014.
=1
(19)
12
11
60
(20)
1.1
0.9
30
20
20 houses
50 houses
75 houses
150 houses
10
TABLE II.
0.7
0.6
0.5
Aggregation
level
# of
Houses
1.
2.
3.
4.
20
50
75
150
1
minute
18.0
12.6
10.6
7.7
IV.
0.4
10
20
30
40
Sampling Interval (Minutes)
50
60
=1
( )
(21)
Eq. (20) is applied on the selected data set and all the
results are shown in Fig. 5 and Table II.
TABLE I. ARSD FOR DIFFERENT COMBINATIONS OF AGGREGATION LEVELS
AND SAMPLING INTERVALS
Aggregation
level
# of
houses
1.
2.
3.
4.
20
50
75
150
1
minute
1.086
0.900
0.855
0.803
15
minutes
0.805
0.753
0.738
0.690
30
minutes
0.761
0.681
0.628
0.554
15
minutes
42.2
29.7
25.5
21.4
30
minutes
44.6
35.1
33.2
29.9
60
minutes
56.6
51.0
49.8
47.7
DISCUSSION
1001
60
50
40
30
20
Sampling Interval (Minutes)
10
0.8
:
1 :
:
40
20 houses
50 houses
75 houses
150 houses
50
60
minutes
0.651
0.547
0.496
0.424
211
17th IEEE Mediterranean Electrotechnical Conference, Beirut, Lebanon, 13-16 April 2014.
[4]
[5]
[6]
[7]
[8]
[9]
CONCLUSIONS
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
REFERENCES
[1]
[2]
[3]
[22]
[23]
212