Académique Documents
Professionnel Documents
Culture Documents
Acknowledgement ..........................................................................................................................1
Abstract ...........................................................................................................................................2
Chapter 1
Introduction ....................................................................................................................................4
1.1 Introduction ............................................................................................................................4
1.2 Background of Study ..............................................................................................................4
1.3 Problem Statement .................................................................................................................5
1.4 Objective of Project ................................................................................................................5
1.5 Scope of Study .......................................................................................................................5
Chapter 2
Literature Review ..........................................................................................................................7
2.1 Introduction ............................................................................................................................7
2.2 Journal and Research Papers ..................................................................................................7
Chapter 3
Methodology .................................................................................................................................10
3.1 Introduction ..........................................................................................................................10
3.2 Load Analysis .......................................................................................................................10
3.3 System Design ......................................................................................................................16
3.4 Fuzzification .........................................................................................................................17
3.5 Fuzzy Inference Rule ...........................................................................................................20
3.6 Defuzzification .....................................................................................................................23
Chapter 4
Results and Discussions ...............................................................................................................31
Chapter 5
Conclusion ....................................................................................................................................35
References .....................................................................................................................................36
ACKNOWLEDGEMENT
First of all, I would like to say Alhamdulillah, praise to our God Allah S.W.T for the
blessings and giving us a good health to finally able to accomplish this report. I would also like
to express our most sincere gratitude to our dedicated lecturer, Datin Prof. Ir. Dr. Shah Rizam
Mohd Shah Baki for all her guidance, patience and also for being very supportive for me to
finish the report perfectly to fit the specifications and requirements to finally able to make this
report almost perfect and flawless. Also being very supportive in all possible ways, my parents
are not to be forgotten as they bless us with unstopped prayers, allowing us to think clearly and
work effectively from the very beginning to the end, without any loose ends. Last but not least, I
would also like to thank all of our friends, especially from the faculty of engineering that spent
their time commenting and contributing brilliant ideas which is vital to the outcome of this
report, and for their outstanding commitment and support.
Abstract
Load forecasting has been always been used by power electrical provider to increase
efficiency as well as to meet demand especially during peak hours. This action can prevent
downtime and power outages in the most demanding hours and several methods have been
practiced to forecast load such as expert system and statistical model-based learning. The fuzzy
logic approach is used in this report to achieve similar result. Past load data ranging from
September to October 1992 is analyzed to develop this system. Peak load of one day is defined at
two times, which are 11 am in the morning and 3 pm in the afternoon. The fuzzy logic system
inputs will be consisting of previous two peak loads and day type. A rule block is used to give
result in the forecasted peak load for next day. The Forecasted load output is expected to contain
less than 2% error when compared to the actual value.
CHAPTER 1
CHAPTER 1
INTRODUCTION
This study will forecast peak load for the next day based on past peak load values and
type of day. This is a type of short term load forecasting that also includes forecasting in
terms of weekly, daily and hourly. This data is important for electrical provider companies,
to generate enough electricity in peak hours. Peak load of a given day is observed from past
load data. The finding from past load data shows peak load occurred two times in a day,
which is on 11 am and 3 pm on weekdays. Most of Saturdays are half-day working day,
which gives the same load profile as weekdays before noon. After office hours at noon, the
load graph descends similar to holidays. On Sundays, the load profile is low throughout the
day and the peak load on Sundays is observed to be occurred at night at around 8 pm.
This fuzzy logic system will be simulated on FuzzyTech 5.54. The analyzed data from
load curve analysis is used to assign value for fuzzification, which is a process to convert
numerical value to linguistic variable. These inputs in linguistic variable will be used to
assign fuzzy rules that are important to make decision. The final value will be defuzzified
using the best method that gives least error to ensure the reliability and trustworthiness of the
system.
During generation and distribution process in electrical power provider operation, the
ability to forecast peak load in a day can greatly improve efficiency and save cost. Regular
maintenance and repairs can be done when the electricity demand is low and the provider can
be ready to generate more during peak hours, but accordingly. A system must be developed
to forecast peak load in a day and taking in consideration environmental factors as well as the
type of day.
The study is based on raw electrical load data from September to October 1992, which
the data is used to design a fuzzy logic system to forecast peak load of a given day, based on
the peak load of previous day and peak load of the same day on last week. The accuracy of
forecasted load is determined to determine whether the system is capable or not to be
implemented on the real world.
CHAPTER 2
CHAPTER 2
LITERATURE REVIEW
1.1 INTRODUCTION
Academicians and electrical engineers around the world have already applied the
artificial intelligence techniques to forecast electrical load. Some of these
accomplishments are already published in IEEE Library Database. A few papers are
found relevant in assisting in the completion of this report. The summary of these papers
are as follows.
Load forecasting with fuzzy logic has been applied worldwide especially in
countries with four seasons, to ensure high efficiency in generation and transmission [1].
Fuzzy logic system in real world application also includes expert system and neural
network to assist in achieving higher accuracy in forecasting. Previous load data is
evidently useful to teach these systems to predict peak load in short term forecasting.
Fuzzy values (linguistic variable) can increase accuracy of a system as numerical method
can largely affected by any value discrepancies. Large or small geographical area also
can contribute to load forecasting as proven by data in Midwest US and the UK [2]. In
[3], type-2 fuzzy logic system is used for load forecasting, which type-2 fuzzy logic
system consist of two steps which is reducer block and then defuzzifier whereas type-1
fuzzy system only has defuzzifier.
It is also woth mentioning that Saleh Ahmadi in [5] presents Iran power
generation and specifically study the Sanandaj Power Network. The case study lists the
factors that is considered during load forecasting including economy and power
disturbance. In [6] Amit Jain in India shows how temperature and humidity can affect
7
load demand in tropical countries. Average day temperature is plotted against average
load, the same is done to average humidity. M.F.I Khamis in [8] shows fuzzy logic
application for a small scale power system. The proposed model can be used to forecast
load one week ahead, every single day in a week is forecasted. The system also
incorporates Data Dynamic Exchange Server to obtain previous load data. In [10] A.R
Koushki simulates that the combination of fuzzy logic and neural network will result in
much more accurate forecasts. The learning algorithm mentioned in this paper is Locally
Linear Model Tree (LoLiMoT) and the results are compared to multilayer perception and
Kohonen Classification and Intervention Analysis.
CHAPTER 3
CHAPTER 3
METHODOLOGY
3.1 INTRODUCTION
Raw data obtained must be analyzed and interpreted before decision is made in designing
a fuzzy logic system. The fuzzy logic system design will follow a project flow chart and
become a complete system once all the processes are done.
10
The first week load graph is shown in Figure 1. The load reduces significantly during
Sunday and is also low during Saturday. For the rest of week, the load curve is quite
consistent, except for a drop on Monday at 8 pm.
11
For the third week of September load curve, the lowest load reading is during Sunday as
illustrated in Figure 3. On Saturday, offices are opened half-day, supporting that the curve is
the same with other weekdays curves before noon and started to fall afterwards.
12
A major power outage occurred on 29th September, causing the load curve to fall greatly
on Tuesday as shown in Figure 4. Recovery process continues in the next day, showing low
load curve (green) similar comparable to Sundays curve (orange). The data collected during
power outage may not be used in actual forecasting system as a reliable system must use
stable, regular load curve on a normal day.
The pattern goes on for the first week of October. There are no irregular values during the
weekdays which showing normal load curve and the load curve for Sunday is the lowest of
the week as usual. On Saturday which is a half-day work day, the load profile decreases after
noon when offices closed for the day as shown in Figure 5.
13
On second week of October, the load profile for weekdays follow usual trend, except for
Tuesday which shows significant drop from 11 pm to 12 am but the load recover the next hour as
usual as shown in Figure 6. This occurrence must be caused by typing error in the data input
process or a small and short power outage. The load curve for weekend also follow the usual
trend, low throughout the day on Sunday and half-day working load curve for Saturday.
14
For the third week of October, weekdays load plot shows regular pattern except for Friday
that has a slight drop at 3 pm as shown in Figure 7. Sunday load profile shows usual low pattern
and Saturday plot also dropped after office hours ended at noon.
On the fourth week of October, the load graph suggested that it was public holiday on Monday,
which shows similar load profile to Sunday as shown in Figure 8. Other weekdays show regular
pattern of load plot and plot on Saturday decreases after noon.
16
Start
Yes
Large error between
forecasted and actual load?
No
Accept system and analyze results
End
The block diagram of this system is represented in Figure 10. Blocks on the left side
represent input of this sytem and block on the right hand side represents output of the system.
The block in the middle is rule block, where fuzzy rules is located and fuzzy operations are
performed.
17
3.3 FUZZIFICATION
The system consists of 5 inputs which are Day, LastDayPeak1, LastDayPeak2,
LastWeekPeak1 and LastWeekPeak2. These inputs must be converted from numerical values to
linguistic variables. Assignment of fuzzy membership for power values can be equal for all the
inputs. This is to ensure consistent result for all the inputs. Although the memberships have equal
value, the name for each membership function must be different from each other. This is
important to prevent confusion during defining fuzzy rules process.
The first input used is day, numerically written from 1 to 7 where 1 is Monday and 7 is
Sunday. The membership function is as shown below in Figure 11
The next four inputs of the fuzzy logic system are various load, which are LastDayPeak1,
LastDayPeak2, LastWeekPeak1, and LastDayPeak2. LastDayPeak1 is the peak load of previous
day at 11 am, and the membership function is shown in Figure 12.
LastDayPeak2 describes peak load from previous day at 3 pm which when second peak occurred
in a day and the membership function is illustrated in Figure 13.
19
LastWeekPeak1 is the peak load of same day on previous week at 11 am, and the membership
function is shown in Figure 114.
20
21
23
Fuzzy inference chosen to be used in this system is Max-Min method. For this method,
the membership values in the two or more conditions within IF statement in a fuzzy rule is
compared. Minimum value from these compared value is taken as AND operation is used in IF
statement. For example in Figure 20, the first two membership functions shows the fuzzy value
for two items in the IF statement. The third membership figure shows the resulting fuzzy
inference by using Max-Min method.
3.6 DEFUZZIFICATION
The last step in completing this fuzzy logic system defuzzification method. Centre-ofMaximum (CoM) method uses all maximum values in fuzzy inference membership, and all of
these maximum points are multiplied with its respective crisp value. The summation of all these
terms are then being divided by sums of all membership value of maximum points. This method
of calculation is represented by this formula in Figure 21 where is membership value of all
maximum points and Y is crisp value in x-axis.
=
( )
24
Mean-of-Maximum defuzzification uses the edge value in highest point obtained in Max-Min
fuzzy inference and divide by two to obtain crisp value, as shown in Figure 22.
() =
a+b
2
()
()
First peak for Monday which is the day before is given as 4362.8 MW is a member of fuzzy set
LDayHigh to the degree of 1.00 as shown in Figure 25.
26
First peak for Tuesday which is the week before is given as 4175.1 MW is a member of fuzzy set
LWeekMedHigh to the degree of 0.13 and LWeekHigh to the degree of 0.99 as shown in Figure
27.
Performing Mamdani Fuzzy Inference (Max-Min Method) will result in a plot as follows in
Figure 29.
Centre-of-Maximum (CoM)
Simulating CoM defuzzification gives the following result as in Figure 30.
28
Besides from using software simulation to find defuzzified crisp value, manual
calculations can be performed and compared.
=
=
( )
(ii)
Mean-of-Maximum (MoM)
Simulating CoM defuzzification gives the following result as in Figure 31.
29
(iii)
Centre-of-Area (CoA)
Simulating CoA defuzzification gives the following result as in Figure 32
()
()
30
CHAPTER 4
31
CHAPTER 4
4.1
INTRODUCTION
In this chapter, result is collected and represented in table and graph form. Three
defuzzification methods are used in this chapter. The values from actual and forecasted load are
compared and error less than 2% is expected to conclude that the method is the most suitable to
be implemented.
4.2
in Table 1. Three defuzzification methods (Mean of Maximum, Centre of Maximum and Centre
of Area) are performed and compared to the actual values. A graph is also generated in Figure 33
to graphically show differences in simulated and actual values.
Day
Actual Peak
Load
MoM
CoM
CoA
Thursday
4395.2
3976.7
4325.4
4325.9
Friday
4371.2
3976.7
4247.2
4250.1
Saturday
4213.0
3976.7
4172.4
4169.8
Sunday
3209.0
3976.7
3097.3
3102.4
Monday
4443.5
3976.7
4303.0
4411.0
Tuesday
4442.5
3976.7
4315.5
4310.1
Wednesday
4439.0
3976.7
4307.3
4308.1
Defuzzification Comparison
5000
4500
4000
3500
3000
2500
2000
1500
1000
500
0
0
Day
Actual Load
MoM
CoM
CoA
Based on Table 1, Relative error is calculated using formula as follows and the results are
tabulated in Table 2.
(%) =
Day
| |
100
Forecasted Peak Load (MW)
Actual Peak
Load
MoM
CoM
CoA
Thursday
4395.2
9.5
1.59
1.57
Friday
4371.2
9.02
2.8
2.7
Saturday
4213.0
5.6
0.96
1.02
Sunday
3209.0
23.9
3.5
3.3
Monday
4443.5
10.5
3.1
0.73
Tuesday
4442.5
10.4
2.9
2.9
Wednesday
4439.0
10.4
2.9
2.9
By comparing these graphs of defuzzification methods being used in all of the data, Centre-ofArea (CoA) and Centre-of-Mass (CoM) method has little difference in result. Between CoA and
CoM methods, CoA methods is more accurate to be used as this method takes more sample in
calculation to come up with crisp defuzzified value contrary to CoM method that take only one
reading per membership value in Max-Min fuzzy inference diagram. Therefore, CoA method is
chosen to be the best method to implement this system.
34
CHAPTER 5
35
CHAPTER 5
CONCLUSION
The system can successfully forecast peak load based on previous peak load data and
type and day. The best implementation is also determined which is using Centre-of-Area (CoA)
defuzzification which takes samples from more points than two other defuzzification method.
The relative error margin is said to be kept under 2% but it cannot be achieved for all instance
despite countless trial and error in modifying the fuzzy rules. The final system managed to
achieve less than 2% error in certain circumstances.
To improve this system in the future, fuzzy logic can be used alongside other artificial
intelligence techniques such as artificial neural network or particle swarm optimization. This
approach can greatly improve the systems accuracy and reliability. Another suggestion is to use
many rule blocks and more inputs that takes into account environmental factors such as weather
and seasons.
36
REFERENCES
1. Sachdeva, S.; Verma, C.M., "Load Forecasting using Fuzzy Methods," Power System
Technology and IEEE Power India Conference, 2008. POWERCON 2008. Joint
International Conference on , vol., no., pp.1,4, 12-15 Oct. 2008.
2. Shu Fan; Yuan-Kang Wu; Wei-jen Lee; Ching-Yin Lee, "Comparative study on load
forecasting technologies for different geographical distributed loads," Power and Energy
Society General Meeting, 2011 IEEE , vol., no., pp.1,8, 24-29 July 2011.
3. Kurniawan, T.Y., "Electrical load time series data forecasting using interval type-2 fuzzy
logic system," Computer Science and Information Technology (ICCSIT), 2010 3rd IEEE
International Conference on , vol.5, no., pp.527,531, 9-11 July 2010
4. Xiaojuan Liu; Enjian Bai; Jian'an Fang; Lunhan Luo, "Time-variant slide fuzzy timeseries method for short-term load forecasting," Intelligent Computing and Intelligent
Systems (ICIS), 2010 IEEE International Conference on , vol.1, no., pp.65,68, 29-31 Oct.
2010
5. Ahmadi, S.; Bevrani, H.; Jannaty, H., "A fuzzy inference model for short-term load
forecasting," Renewable Energy and Distributed Generation (ICREDG), 2012 Second
Iranian Conference on , vol., no., pp.39,44, 6-8 March 2012
6. Jain, A.; Srinivas, E.; Rauta, R., "Short term load forecasting using fuzzy adaptive
inference and similarity," Nature & Biologically Inspired Computing, 2009. NaBIC 2009.
World Congress on , vol., no., pp.1743,1748, 9-11 Dec. 2009
7. Wang Xiao-Wen, "The short-term load forecasting based on fuzzy inference rules,"
Electricity Distribution (CICED), 2012 China International Conference on , vol., no.,
pp.1,3, 10-14 Sept. 2012
8. Khamis, M. F I; Baharudin, Z.; Hamid, N. H.; Abdullah, M. F.; Nordin, F. T., "Short
term load forecasting for small scale power system using fuzzy logic," Modeling,
Simulation and Applied Optimization (ICMSAO), 2011 4th International Conference on ,
vol., no., pp.1,5, 19-21 April 2011
37
38