Professional Documents
Culture Documents
IMPORTANT NOTE: Power Analytics Corporation’s software products are tools intended to be
used by trained professionals only. They are not substitutes for your professional judgment or
for independent verification and testing of results as they pertain to your specific application.
Use of all Power Analytics Corporation software products is governed by the terms and
conditions of the End-User License Agreement (“EULA”) you accepted when purchasing and
installing the software. You must comply with these terms and conditions in applying the
instructional material in this manual. If you do not have or are unfamiliar with the contents of
your EULA for this software, you should request, read, and understand a copy of your EULA
before proceeding.
Please accept and respect the fact that Power Analytics Corporation has enabled you to make an
authorized disk as a backup to prevent losing the contents that might occur to your original disk drive. DO
NOT sell, lend, lease, give, rent or otherwise distribute Power Analytics programs / User's Guides to
anyone without prior written permission from Power Analytics Corporation.
i
Distribution Load Forecast Program
TABLE OF CONTENTS
1. Introduction ........................................................................................................................ 1
ii
Distribution Load Forecast Program
TABLES
Note: You can view this manual on your CD as an Adobe Acrobat PDF file. The file name is:
You will find the Test/Job files used in this tutorial in the following location:
iii
Distribution Load Forecast Program
1. Introduction
The task of distribution system planning involves determining sizes, locations and timing of future
changes, such as additions of system components (substations, lines, feeders, etc.). Planning of
distribution systems requires forecast of the future load demands in terms of these components.
The quality and accuracy of system planning depends on the quality and accuracy of the data and
load forecast.
Geographic location of loads is analyzed using a small area forecasting approach, which divides
the utility service area into a number of small areas and forecasts the load in each one, thus
determining where and how much load there will be. There are two methods to divide a system
into small areas:
Each method of small area definition has advantages and disadvantages. The grid-based
methodology requires considerable input data, including not only historical load records in each
grid block, but also economic, social, demographic and land-use information, to provide accurate
results. For most utilities, it is difficult to make these data available. The Power Analytics
DesignBase distribution load forecast program uses the first method (i.e. dividing the utility service
area into feeders, substations, or zones, defined by equipment). The major merit of this
methodology is that it requires only historical load data for a few years, which are readily available
to any utility. The load increase or decrease limit should be evaluated with regard to other
important elements, such as availability of land, water, as well as social and economical factors,
that could impact the load forecast trend.
(3) Stochastic Time Series. In many cases, this technique can provide more
accurate results than the multi-regression methods. In particular, this
technique can conduct not only yearly, but also monthly, load forecasting.
It is important to appreciate that some differences always exist between the forecast and reality,
no matter which method is used. The multi-regression method may be better for some types of
historical data, while the time series technique is better for others. It is, therefore, suggested to
use both and compare the results whenever possible. Forecasting accuracy is a statistical
1
Distribution Load Forecast Program
concept. In other words, a high average of accuracy in load forecasting for sets of distribution
components (substations, feeders or zones) and many points of time (years or months), does not
necessarily mean a small error for every component and every time period (for example each
year) to be forecasted.
2
Distribution Load Forecast Program
2.1 Multi-Regression
The multi-regression technique is based on curve fitting to historical load data for each distribution
component (small area).
where
L i t - estimate of load for component i for year t
t - Year No.
Li (t) = b0 + b1 x1 + b2 x2 + b3 x3 + b4 x4 + b5 x5 + b6 x6 (2)
Assuming that N sets of values for x1 , x 2 , x 3 , x 4 , x5 , x 6 , L i t corresponding to N year of historical load
records are expressed by:
x k1 , x k2 , x k3 , x k4 , x k5 , x k6 , y k (k=1,2,....N}
Applying the least square error estimate method, the following normal equations can be obtained:
S 11 b1 + S 12 b2 + .... + S 16 b6 = S 1y
S 21 b1 + S 22 b2 + .... + S 26 b6 = S 2y
(3)
............
S 61 b1 + S 62 b2 + .... + S 66 b6 + S 6y
The coefficients b 0 , b1 , b 2 , b 3 , b 4 , b5 , b 6 can be obtained by solving Equation (3). Once the b coefficients
are obtained, load forecasting can be conducted by substituting the number of years to be forecasted into
Equation (1). Large inaccuracy may occur in some cases, because of the difference between the historical
data trending and future development trending. Horizon year load estimates can stabilize trending to an
extent where even mediocre estimates are worth using. The horizon year estimates add a load estimate at the
horizon year, which is treated in the same way as a historical load in the above multi-regression
calculations.
3
Distribution Load Forecast Program
The main disadvantages of the conventional multi-regression method are: (a) The normal
equation may become ill-conditioned in some cases, and (b) it cannot identify important
regression variables. Both of these items can lead to large inaccuracies in calculations.
The successive multi-regression method used in the program includes the following
improvements:
(1) Correlation coefficients rij are used to replace the S ij in Equation (3) leading to a
high accuracy in resolution. rij is defined as:
sij
r ij = (4)
sii s jj
The historical load data may have been contaminated due to the switching of loads between two
substations or feeders. A load may be transferred from one substation or feeder to another for
some reason. For example, a load may be transferred during construction or in the case where a
planned new or larger transformer is delayed.
The coupling multi-regression technique is used to eliminate the effects of load transfers between
two substations or feeders. A weighting least square error method is applied to replace the
conventional least square error approach. The same weighting is put on the square error terms
corresponding to years where load transfers take place between the two substations or feeders.
This technique does not require the magnitude and direction of load transfers as input data.
Additional required data are only with regard to which year(s) a load transfer took place.
(1) Historical load data can be viewed as a discrete time series that is not stationary and
contains a growth trending (for yearly load records) or periodic variations (for monthly
load records). The growth trending and periodic variation can be filtered by using a
forward differencing technique to obtain a stationary time series.
x1 , x 2 , x 3 , x 4 ,...... x N
4
Distribution Load Forecast Program
where:
xt = xt -
*
(6)
The p i in equation (5) are called partial auto-correlation coefficients and can be estimated by using
the Yule-Walker equation:
p1 + r 1 p2 + r 2 p3 + .... + r m-1 p m = r 1
r 1 p1 + p2 + r 1 P3 + .... + r m-2 p m = r 2
(7)
..............................
where ri are called auto-correlation coefficients and can be estimated using the equations:
N
1
x =
N
xt
t=1
N -k
1
ck =
N
( x t - x )( x t+k - x )
t=1
(8)
r k = c k / c0
(k = 0,1,2,...,m;m » N)
(3) After p i are calculated using Equation (7), the forecasting can be conducted by the forecast
equation:
* * *
x N+L = p1 x N+L-1 + ..... pm x N+L-m (9)
The forecasted results from Equation (9) are those corresponding to the stationary time
series. Forecasting for the original discrete load time series can be obtained by using a
backward differencing technique.
5
Distribution Load Forecast Program
1. Zone(s)
2. Substation(s)
3. Feeder(s)
4. Two substations with load transfer
5. Two feeders with load transfer
Note: If the object is zone(s), substation(s) or feeder(s), either of the two methods
can be chosen. If the object is two substations or feeders with a load
transfer, only the multi-regression method can be used, and this is
automatically carried out by the program.
1. KW
2. KVA
3. MWh
1. Yearly forecast
2. Monthly forecast
Note: If yearly forecast is selected, either of the two methods can be chosen. If
monthly forecast is selected, only the time series method can be used, and
this is automatically carried out by the program.
6
Distribution Load Forecast Program
(1) Users should input basic historical load data, including year number (i.e. 1990), for yearly forecasting,
or year-month number (i.e. 1990.04), for monthly forecasting and load records. At least five years
of load records are required for yearly forecasting, and, in the case of monthly
forecasting, at least 36 months of load records are required, and the number of the
load records should be a multiple of 12. There is no limit to the number of historical load
records that can be entered. However, extensive amounts of past history is not suggested, because
future load trending is mainly affected by recent load development modes, and much
less by distant past load history.
(2) If the multi-regression method is chosen, horizon year load estimate data are required. The horizon
year should generally be within 3 to 10 years after the last year to be forecasted, and
load estimates at the horizon year may be a reflection of economic, financial or environmental limits on
load growth. If providing horizon year load estimates is difficult, the time series method can be used to
obtain forecasting results first. Then these results can be used as a reference for the horizon estimates.
(3) To conduct load forecasting for two substations or feeders involved in a load transfer,
additional data are required. For the MWh forecasting, information regarding in which year(s)
a load transfer took place is required, and, for KW or KVA forecasting, information regarding
in which years a load transfer only took place in the peak period is required. The program
does not need to input magnitudes nor directions of load transfers.
(4) The maximum number of zones, substations or feeders in a single run is 40. When the
number is more than 40, load forecasting should be done in two or more runs.
(1) There is no limit to the number of years to be forecasted. However, the accuracy of results for
very long term forecasting decreases.
(2) Monthly forecasting only provides results for the 12 months immediately following the last year
of historical data.
7
Distribution Load Forecast Program
8
Distribution Load Forecast Program
d) Click Save button from New Job File dialog box to create entry data view.
9
Distribution Load Forecast Program
e) Choose an object (e.g. Substations) from the data entry dialog box as follows:
10
Distribution Load Forecast Program
Select the appropriate settings in the Distribution Load Forecasting Configuration dialog box and click
OK to save. This includes:
Enter description information and forecasting length. The years 1989 through 1995 are
automatically entered for historical data. For each year, enter the load history for each unit.
If the multi-regression method is used, enter a horizon year in the “Horizon” block, and load
estimates for each unit in the block at the bottom right corner of the dialog box.
11
Distribution Load Forecast Program
There are twenty substations in the job file shown above, and the historical data are for the
seven years from 1989 to 1995. The forecasting length is ten years. The horizon year is 2012.
12
Distribution Load Forecast Program
If an object of “2 Feeders with LT” or “2 Substations with LT” is chosen, specify the year(s) (at
least one year) in which a load transfer took place by clicking in the “load transfer” block located
in the bottom left corner of the dialog box, as shown below:
13
Distribution Load Forecast Program
1. After settings are entered in the Distribution Load Forecasting Configuration dialog box, and the
OK button is clicked, options of objects can no longer be changed from the data entry dialog box. To
make changes to objects, the From or To year field must be cleared in the Configuration dialog box.
2. If the From or To year field is cleared, all historical data in the data entry view will also be
cleared, and must be re-entered.
3. After the historical data have been entered, caution must be used when changing
from the “Yearly” to the “Monthly” option, or vice versa. The data for yearly forecasting are
of a completely different nature, and limits different from the monthly forecasting are used.
Therefore, such changes may lead to lost or damaged data.
After starting the program, either create a new job file, by selecting the File/New command, or
open an existing job file, by selecting the File/Open command.
Creating or opening a file opens the dialog box shown below, which contains the settings for data
entry. This view will be blank when creating a new file.
Results of the Load Forecast computations are generated by clicking the Run or Calculate
command, and may be viewed by selecting the View/Report or View/Output Results command.
14
Distribution Load Forecast Program
The following list provides an overview of commands from the File menu:
New The File/New command displays the New Job File dialog box, and
creates a blank view of the file.
Open The File/Open command displays the Open Job File dialog box for
selection of an existing LFC job file. The default directory for opening
existing files is defined in the Job Files section of the DesignBase
Directories and Packages dialog box, which is accessed using the
Options/Directories and Packages command.
Close The File/Close command checks for unsaved changes to the active job
file. If there are unsaved changes, the user is prompted to save or
discard the changes, and all dialog boxes for the file are closed.
Save The File/Save command saves the open job file. If the file is unnamed,
the File/Save As dialog box is displayed for entry of the file name and
indication of where the file is to be saved.
Exit The File/Exit command closes the Distribution Load Forecast program.
If there are unsaved changes to the job file, the user is prompted to
save or discard the changes, or cancel the Exit command.
15
Distribution Load Forecast Program
The File\Open\Samples\LdFrcast\Exam1 command opens the following data entry dialog box:
Click the Run button or select the Calculate command to calculate the Load Forecasting results.
Click the Error Info button to check for valid data entry and successful computation.
16
Distribution Load Forecast Program
Users can select Input Data, Output Results, Graph or Report from View menu at any time:
If a view is already open, the command will be marked with a check symbol (i.e. Input Data).
17
Distribution Load Forecast Program
View/Input Data
The View/Input Data command opens the following Input Data dialog box. Select either
Summary or Loading to display different input data. Click the Refresh button to view any
changes to the input data. Click the Close button to close this view.
18
Distribution Load Forecast Program
View/Output Results
The View/Output Results command opens the following results dialog box. Click the Refresh
button to re-calculate and view new results for changes in input data. Click the Close button to
close this view.
19
Distribution Load Forecast Program
View/Graph
20
Distribution Load Forecast Program
In this view which curves are displayed can be determined by selecting the desired curves and clicking the Sel
button on right-toolbar. A hard copy of the graph can be printed using the Output menu. Select View/Display or
click the Display icon to display right-toolbar if it has been closed.
View/Reports
21
Distribution Load Forecast Program
The View/Reports command formats the input data and results into the report shown below:
22
Distribution Load Forecast Program
After the data entry for all settings has been completed, selecting the Calculate command, or
clicking the Run button in the data entry dialog box, obtains the Load Forecast results.
23
Distribution Load Forecast Program
The Windows commands provide control over the arrangement of the calculation dialog boxes
within the application window area.
Arrange Icons The Window/Arrange Icons aligns the icons for any minimized
windows (see System Menu) in a row at the bottom left hand side of
the application window area.
Close All The Window/Close All command closes all dialog boxes. This is
equivalent to selecting the File/Close command.
1 (etc.) Each open dialog box will have an entry in the Window menu.
Selecting one of these entries will display that dialog box in the
application area.
24
Distribution Load Forecast Program
Help commands provide access to on-line information designed to assist the user of the
Distribution Load Forecasting program.
Contents Displays the master table of contents for the Help system.
Using Help Displays general information about how to use the Help system.
25
Distribution Load Forecast Program
When the mouse is used to move the cursor to an icon on the toolbar, the command text label is
displayed under the icon.
New File
Open File
Input
Graph
Report
Help
CODE 4: When monthly type is chosen, the number of historical load records should be a
multiple of 12
CODE 5: Horizon year should be within 3 to 10 years after the last year forecasted
CODE 7: Invalid data load in unit XX and at year XX (the unit and year number are from the error
struct)
26
Distribution Load Forecast Program
7.1 Example 1
This example shows the yearly KW load forecasting of 20 substations for the 10 years from 1996
to 2005. The historical KW load data for the 7 years from 1989 to 1995 are given as input.
Checked by : Date:
27
Distribution Load Forecast Program
HISTORICAL LOAD DATA (KW )
---------------------------------------------------------
YEAR Subst11 Subst12 Subst13 Subst14 Subst15
---------------------------------------------------------
1989 11494 2334 1297 4074 14184
1990 12097 2434 1168 4336 14943
1991 12087 2414 1317 4616 15940
1992 13109 2305 1217 4486 16429
1993 11896 2334 1297 4729 16489
1994 12957 2385 1397 4747 16569
1995 12679 2424 1406 5457 16419
---------------------------------------------------------
28
Distribution Load Forecast Program
1996 12968 2440 1438 5610 16736
1997 13252 2455 1457 5970 17412
1998 13341 2470 1462 6011 18107
1999 13603 2485 1491 6208 18600
2000 13783 2500 1504 6428 18809
2001 13973 2515 1521 6700 18922
2002 14183 2530 1544 6986 19169
2003 14372 2545 1557 7204 19620
2004 14573 2560 1578 7427 20178
2005 14771 2575 1596 7625 20674
---------------------------------------------------------
29
Distribution Load Forecast Program
Figure 1
30
Distribution Load Forecast Program
31
Distribution Load Forecast Program
32
Distribution Load Forecast Program
(a) The comparisons between the results obtained using the two forecasting techniques
and the actual loads for the 20 substations for 1996 are summarized in Table 1.
-------------------------------------------------------------
Subst. Actual Load Time Series Multi-regression
(KW) (KW) Diff(%) (KW) Diff(%)
-------------------------------------------------------------
Subst1 3387 3179 6.1 3271 3.4
Subst2 12628 12601 0.2 12959 2.6
Subst3 11942 11671 2.3 11269 5.6
Subst4 2090 2175 4.0 2321 11.0
Subst5 9287 9491 2.2 9559 2.9
Subst6 5052 4858 3.8 4934 2.3
Subst7 699 638 8.7 742 6.2
Subst8 8408 8290 1.4 8381 0.3
Subst9 6154 6059 1.5 6001 2.5
Subst10 8462 8962 5.3 9100 7.5
Subst11 12460 12968 4.1 12676 1.7
Subst12 2434 2440 0.2 2393 1.7
Subst13 1446 1438 0.6 1381 4.5
Subst14 5074 5610 10.6 5134 1.2
Subst15 16409 16736 2.0 17093 4.2
Subst16 529 571 7.9 561 6.0
Subst17 16913 18321 8.3 17842 5.5
Subst18 13256 13033 1.7 13400 1.2
Subst19 4529 4315 4.7 4336 4.3
Subst20 3281 3252 0.9 3322 1.2
-------------------------------------------------------------
(b) Most of the substation loads have growth trending in the 10 year forecasting.
However, there are three substations (4, 7 and 16) whose loads show a decreasing
trend, and one substation (6) has a flat load trending.
33
Distribution Load Forecast Program
7.2 Example 2
This example shows the yearly MWh forecasting of 12 zones for the 10 years from 1996 to 2005.
The historical MWh data for the 9 years from 1987 to 1995 are given as input.
Checked by : Date:
34
Distribution Load Forecast Program
35
Distribution Load Forecast Program
Figure 2
36
Distribution Load Forecast Program
37
Distribution Load Forecast Program
---------------------------------------------------------
YEAR Zone11 Zone12
---------------------------------------------------------
1987 321945 319706
1988 326156 351066
1989 368854 377184
1990 386381 390208
1991 392742 393532
1992 373208 408273
1993 396705 387546
1994 405072 417678
1995 405650 401560
---------------------------------------------------------
Horizon estimates:
2012 580000 580000
---------------------------------------------------------
38
Distribution Load Forecast Program
The results obtained using the two forecasting techniques and the actual MWh for the
12 zones for 1996 are summarized in Table 2.
-------------------------------------------------------------
Zone Actual Load Time Series Multi-regression
(MWh) (MWh) Diff(%) (MWh) Diff(%)
-------------------------------------------------------------
Zone1 355562 384910 8.3 385427 8.4
Zone2 350112 367427 4.9 370863 5.9
Zone3 339605 344822 1.5 350201 3.1
Zone4 350760 361354 3.0 360206 2.7
Zone5 357967 362205 1.2 371598 3.8
Zone6 349928 357368 2.1 360814 3.1
Zone7 393678 385413 2.1 399977 1.6
Zone8 437582 433849 0.9 422691 3.4
Zone9 453650 452285 0.3 461167 1.7
Zone10 435126 477126 9.7 473939 8.9
Zone11 386183 417297 8.1 424138 9.8
Zone12 418523 425006 1.5 432359 3.3
--------------------------------------------------------------
39
Distribution Load Forecast Program
8.3 Example 3
This example shows the monthly KVA forecasting of 7 feeders for the 12 months from April,
1996, to March, 1997. The historical load data for the 36 months from April, 1993, to March,
1996, are given as input. Only the stochastic time series technique can be used for monthly
forecasting.
40
Distribution Load Forecast Program
1995.10 2877 3871 2581 1926 2367
1995.11 3144 4385 2690 2095 2379
1995.12 3309 4505 2921 2321 2486
1996.01 3307 4604 2949 2237 2445
1996.02 3415 4990 2920 1953 2372
1996.03 3067 4451 2658 1998 2142
---------------------------------------------------------
41
Distribution Load Forecast Program
---------------------------------------------------------
Y - M Feeder6 Feeder7
---------------------------------------------------------
1996.04 1367 956
1996.05 1284 998
1996.06 1182 905
1996.07 1247 901
1996.08 1281 954
1996.09 1229 918
1996.10 1394 1017
1996.11 1566 1321
1996.12 1695 1253
1997.01 1700 1167
1997.02 1549 1238
1997.03 1501 1044
---------------------------------------------------------
Y = Year
M = Month
42
Distribution Load Forecast Program
Figure 3
8.4 Example 4
The coupling multi-regression technique. (Choosing "two substations with load transfer")
This example shows how the coupling multi-regression technique works for load
forecasting of two substations with load transfers between them. The original data include
the 9 years of load records from 1988 to 1996. It has been assumed that 2000 KW was
transferred from Substation 2 to Substation 1 during the peak period in 1991, and that
3000 KW was transferred in 1993 and 1995. (In the actual forecasting, it is not necessary
to input the magnitude or direction of load transfers.) The forecasting for the 10 years
from 1997 to 2006 was conducted using both the original data and the data contaminated
by the intentional load transfers. The output reports are as follows. It can be seen that the
same forecasting results are obtained for the two sets of data, which indicate that the
effects contaminated by the load transfer have been automatically filtered out.
43
Distribution Load Forecast Program
Checked by : Date:
Graphical output:
44
Distribution Load Forecast Program
Figure 4
45
Distribution Load Forecast Program
Checked by : Date:
46
Distribution Load Forecast Program
8.5 Example 5
The successive multi-regression and the time series techniques (Choosing "substations")
follow.
Load forecasting can also be done using the general successive multi-regression or time
series technique by selecting the "Substations" as the "Object". The output using the two
techniques for both the original data, and for the data contaminated by load transfers, are
as follows. It can be seen that, in the case of the original data, the results obtained using
the two techniques are reasonably close to the results obtained using the coupling multi-
regression technique (see Example 4). In the case of the data with load transfers, the two
non-coupling techniques lead to large errors. However, this does not mean that coupling
multi-regression can replace the other two techniques. In many cases without load
transfers, the time series and successive multi-regression techniques will provide more
accurate forecasting results.
Checked by : Date:
Description : Distribution Load Forecasting Analysis
Exam5a
47
Distribution Load Forecast Program
Graphical output:
Figure 5
48
Distribution Load Forecast Program
49
Distribution Load Forecast Program
50
Distribution Load Forecast Program
51