IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 17, NO.

2, MAY 2002

491

Long-Term Load Forecasting for Fast Developing

Utility Using a Knowledge-Based Expert System
M. S. Kandil, S. M. El-Debeiky, Senior Member, IEEE, and N. E. Hasanien

AbstractThe application of the classical forecasting methods,

when applied to fast developing utility with a period characterized by fast and dynamic changes, are insufficient and may provide an invaluable dimension to the decision making process. In
this paper, a knowledge-based expert system (ES) is implemented
to support the choice of the most suitable load forecasting model for
medium/long term power system planning. In the proposed ES, the
detailed problem statement including forecasting algorithms and
the key variables (electrical and nonelectrical variables) that affect
the demand forecasts are firstly identified. A set of decision rules
relating these variables are then obtained and stored in the knowledge base. Afterwards, the best model that will reflect accurately
the typical system behavior over other models is suggested to produce the annual load forecast. A practical application is given to
demonstrate the usefulness of the developed prototype system.
Index TermsArtificial neural network (ANN), expert systems (ES), fast/normal developing utility, forecasting methods,
long-term load forecast.

I. INTRODUCTION

ONG-TERM load forecasting represents the first step in

developing future generation, transmission, and distribution facilities. Any substantial deviation in the forecast, particularly under the new market structure, will result in either overbuilding of supply facilities, or curtailment of customer demand.
The confidence levels associated with classical forecasting techniques [1], [2] when applied to forecasting problem in mature
and stable utilities are unlikely to be similar to those of dynamic
and fast growing utilities [3], [4]. This is attributed to the differences in the nature of growth, socio-economics conditions, occurrence of special events, extreme climatic conditions, and the
competition in generation due to the deregulation of the electricity sector with possible changes in tariff structures. Under
such conditions, these forecasting techniques are insufficient to
establish demand forecast for medium/long term power system
planning. Consequently, this case requires separate consideration either by pursuing the search for more improvement in the
existing forecasting techniques or establishing another approach
to address the forecasting problem of such systems.
In this paper, the classical forecasting methods are firstly
applied to obtain the long-term peak demand forecasts, for a
typical fast growing utility as well as normal developing system.
Manuscript received March 8, 2000; revised December 4, 2001.
M. S. Kandil is with the Electrical Engineering Department, Mansoura University, Mansoura, Egypt.
S. M. El-Debeiky is with the Electrical Engineering Department, Ain Shams
University, Cairo, Egypt.
N. E. Hasanien is with the Egyptian Electricity Authority, Studies and Researches Department, Cairo, Egypt, and with the Saudi Electricity Company,
Jeddah, Saudi Arabia.
Publisher Item Identifier S 0885-8950(02)03815-4.

A poor performance is observed when such methods are applied to fast developing system, whereas most of these models
are valid when used to produce the forecasts of the normal
developing system. Consequently, an extended logistic model
is developed to reflect the critical forecasting problem in fast
growing areas. Although the developed model gives an accurate
peak demand forecast compared with the classical models, it is
hardly difficult to dependent on single method for producing
the demand of such fast growing and dynamic system. This
is because several important factors related to the cyclic and
dynamic events that contribute significantly to the system peak
is difficult to involve it into the existing forecasting models.
Thus, there is a need to develop a computational tool which
allow one to store the knowledge associated with this problem
along with the mathematical models to support the choice of
the most suitable load forecasting model, for medium/long
term power system planning. Therefore, the implementation
of long-term forecasting strategies using a knowledge-based
expert system (ES) is then presented in this paper. In the proposed expert system, key system variables which have major
effects on system peak are identified based on past planners
experiences. A set of decision rules relating these variables are
then established and stored in the knowledge base to select the
recommended forecasting. A practical application is given to
demonstrate the usefulness of the developed prototype system.
II. IDENTIFICATION OF THE FORECASTING PROBLEM
A common forecasting problem in most fast growing utilities
is the short and insufficient database, unparalleled experience
of fast growth and unavailability of forecasting techniques capable of addressing adequately this particular problem. Moreover, there exists a unique dynamic pattern of the energy and
demand resulting from the interaction of cyclic social events
occurring periodically on lunar and seasonal temperature effect
related to Gregorian calendar. The resulting system load characteristics are complex to analyze and difficult to predict.
The peak demand of the considered system, has grown up at
an average annual rate in excess of 9% in the last ten years. The
system operates in an area where ambient temperatures can exceed 48 C in summer season. The ambient temperature exhibits
a unique pattern every year. The system peak demand is partially
influenced by the temperature, both its magnitude and time of its
occurrence. In addition to the maximum temperatures, there are
other cyclic and dynamic events that contribute significantly to
the system peak demand. The load of these events vary in magnitude and time every Gregorian year and hence their contribution
to the annual system peak demand is dynamic in nature. These

0885-8950/02$17.00 2002 IEEE

492

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 17, NO. 2, MAY 2002

Fig. 1. Mean absolute percent errors associated with the forecasted load of the
fast and normal developing systems from 1994 to 1998.
Fig. 3. Actual peak load versus fitted peak demand of the improved model for
the fast developing system from 1981 to 1993.

COMPARISON

OF

TABLE I
ACTUAL AND ESTIMATED DEMAND USING
PROPOSED MODEL

THE

Fig. 2. Monthly peak demands of the fast developing system from 1986 to
1998.

cyclic variable loads are unlike the static temperature dependent

loads, which occur during a fixed period every Gregorian year.
III. APPLICATION OF CLASSICAL FORECASTING METHODS
The peak demand forecast is obtained for a typical fast developing system and a normal developing system (Egyptian Power
System in this case) as well, by direct application of the classical forecasting methods during the period 1981 to 1993. Fig. 1
shows the mean absolute percent error (MAPE) of the load forecast for both systems compared with the actual demands during
the period 1994 to 1998. Whereas most of the applied models
are valid for the normal developing system, the large error associated with the forecast of the fast developing system, for five
years only, concludes that there is a need to improve the existing forecasting techniques to address the forecasting problem
of such systems. Hence, an improved model of the critical forecasting problem is explained in Section IV.
IV. IMPROVED MODEL OF SYSTEM DYNAMIC LOAD
CHARACTERISTICS
A composite load characteristic of the considered power
system as shown in Fig. 2, is investigated.
Two main load components can be identified. The first component is an intrinsic static temperature-dependent load while
the other is highly unstable dynamic load. The static load is the
summer air-conditioning demand. The peak of this load moves
in harmony with the maximum temperature. The dynamic load
component results from the load contribution of cyclic events
related to the lunar calendar. One important event is the fasting
ritual of Ramadan. During this month, abrupt rise in power demand is customary in the region. The increase in the power demand is attributed to the simultaneous consumption of energy of
consumers during certain periods. As all lunar months move relative to the seasons, the load contribution of this ritual cyclically
moves with respect to the temperature-dependent load. The in-

teraction of the two load components creates highly unstable

load pattern.
A modified logistic model is suggested to represent the actual
peak demands recorded from 1981 to 1993 as
(1)
And the dummy variable

is given by
(2)

is the annual peak demand, MW,

is a variable
where
comprising the effect of time, maximum temperature and Raand
madan effects on peak demand, is the time (years),
are weighting factors representing the effect of Ramadan
and maximum temperature on the peak demand, respectively,
and , , , , and are constant coefficients.
and
in the period 1981
The weighting factors of both
to 1993 are firstly produced. Then, estimated values of the proposed model parameters are then obtained using the nonlinear
regression [5]. A comparison between actual and fitted peak demands is shown in Fig. 3.
Table I shows peak demand forecasts compared with recorded
demands during the period 19941998.
The relatively large error associated with 1994 is attributed to
the load contribution of the school examinations season, which
was held, for the first time, during the summer. The proposed
model demonstrates better forecasts compared with forecasts
obtained by direct application of classical time-series models.
However, this may not be sufficient, as the future load characteristics will be influenced by many unpredictable events. Such
events are the opening and the closing of the schools, school examinations, rehabilitation of old settlements, and introduction of
competition in generation environment with possible changes in
tariff structures.
The results conclude that it is hardly difficult to depend on
the applications of the classical methods which expect to provide an invaluable dimension to the decision making process.

KANDIL et al.: LONG-TERM LOAD FORECASTING FOR FAST DEVELOPING UTILITY

Fig. 4.

Structure of the expert system for long-term load forecast.

Therefore, computational method which allow one to store the

knowledge based on heuristics and good sense rules alongside
the incorporation of the mathematical forecasting methods, is
essentially needed.
Although the artificial neural network (ANN) approach has
been successfully implemented for short-term load forecasting
[7], there is a difficult to use it in this case where the weights
of many unpredictable variables are a priori unknown. Thus,
a proposed method for combining an expert system with the
mathematical forecasting models is developed to represent the
problem as explained in the next section.
V. EXPERT SYSTEM STRUCTURE
The main components of the proposed expert system are
shown in Fig. 4.
With the knowledge base at hand (rules and facts), an inference engine can be used to search through this knowledge base
according to the solution strategy. The detailed procedures of
the solution strategy to ascertain the accuracy and credibility of
selecting forecasting method using the production rules in the
rule base is described in Section V-D. In addition to knowledge
base, inference engine, solution strategy, a user interface is also
developed in the expert system to facilitate the navigation between the proposed expert system and the user.
A. Knowledge Base
All the knowledge obtained from the past experience, past
studies [1][6] has to be elicited appropriately using an adequate technique. We used the production rules representations.
The data facts (static and dynamic) are included in the knowledge base. The data is represented as a set of variables in FACTS
form, each of which define the value of a given variable as follows.
The variables of the formulated problem can be grouped into
Static and Dynamic Facts as follows:
Static Facts: This kind of knowledge is developed before
starting the planning process. A sample of these facts is: system
conditions to identify the current situation of the system, i.e.,

493

mature, or under developing, isolated or interconnected with

other system, etc.
Forecasting horizon to define the load forecasting period, i.e.,
medium, or long term.

Load pattern to describe the load behavior (stable, or

dynamic pattern, cyclic, or seasonal pattern, or combination of all, time of system peak, load types, etc.)

Historical peak load to indicate annual and seasonal

growth, coincidence factor, area peak, etc.

Historical energy to describe the information related to

number of consumers of each sector, consumption rate,
tariff rate, etc.

Major factors affecting the system peak, i.e., weather

variables, economic variables, demographic variables,
special event, suppressed demand, bulk loads to be
connected into the network, coincidence factor of the
system peak, etc.
The following example of load information shows how these
facts are represented.

>

(deffacts System_Load
(System_peak_load ?spl system_load_factor ?slf
System_coincidence_factor ?scf ) )
(deffacts Areas_Load

>

(area_a p_load ?x area_b p_load ?y area_c p_load

?z))
(assert (System_Load ( ?spl ?slf ?scf)))
(assert (Areas_Load ( ?x ?y ?z)))

>
>

Dynamic Facts: These facts are developed and automatically updated during the inference process to represent the
planning attributes needed for evaluating a decision making
process. Samples of these facts include the following:

load and energy attribute for the estimated load and

energy forecast;

system losses attribute for the estimated system losses;

error attribute related to the forecasting model.

1) Design of Data Structure (Facts): In the developed expert system the facts are stored in objects. The object-oriented
programming (OOP) is used because this structure enables the
use of the notion of hierarchical, inheritance, and data encapsulation. Besides, the possibility of factoring a large program
into smaller units that can be modified independently and more
easily [8], [9].
In an OOP, each object represents a physical object of the
Peak load and energy demand. The objects are defined by a
set of attributes-values, which contains all concrete knowledge
about the particular object. The first step is to identify the relevant object classes of the power system situation and to specify
their relation as shown in Fig. 5
In the next step, common characteristics are generalized to
superclasses, and more specific classes are specialized in subclasses. The results are a hierarchy of inheritance. The classes
IsolLoad, BukLoad, and LodMang are generalized into their superclasses Load which in turn is a specialized of class Electrical
Variables.
2) Design of Rule Base: Rules are expressed as IFTHEN
statement, as shown in the following.
IF (premise)

THEN (conclusion)

494

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 17, NO. 2, MAY 2002

TABLE II
PRODUCTION RULES

Fig. 5.

Object model of demand forecasting of an electrical power system.

In a rule-based system, the domain knowledge is represented as

a set of rules that are checked against a collection of facts or
knowledge about the current situation. When the IF portion of
a rule is satisfied by the facts, the action specified by the THEN
portion is performed.
The rules base stores the called abstract knowledge, which
is composed basically of heuristics of power system demand.
During the inference process, the inference engine tries to match
the facts of the working memory with the knowledge contained
in the rules premises, trying to deduce new facts or to fire actions
into process. In this work, the production rules has a formalism
of object: attribute pairs (refers to properties of an object) as
shown in Table II.
The developed rules are divided into four groups of rules, for
optimizing the inference process as shown in the following.
1) Group 1: Forecasting objective.
2) Group 2: Forecasting techniques.
3) Group 3: Forecasting criteria.
4) Group 4: Forecasting uncertainty factors.
This divisions in-groups allow a more efficient use of information within the entire knowledge base. Each group is composed by several rules. Here is an example of the implemented
rules for selecting an appropriate forecasting technique.
Time series method_Trend Extrapolation
(defrule Trend_Model
(forecast_period ?T)
(utility _developing ?mature ?fast ?normal)
(system_behviour ?stable ?unstable) (system_affected
factors ?defined ?undefined)
(system_trend _pattern ?cyclic ?seasonal ?annual
?all ?)
(system_data ?recent ?minimal ?)
())

Fig. 6.

Solution search strategy of the ES.

(if ( ?T 2 & 5 & normal & stable & defined &annual& resonable & minimal &) then
(assert (time series method))
else
( ?T 2 & 5 & fast & unstable & )
then
(assert (.)))

B. Inference Engine
The inference engine is to link the rules given in the knowledge base and the associated condition input by the user with
data stored in the database in order to solve the problem. The
shell adopted for building the expert system CLIPS [8] allows
the use of production rules and facts pattern matching to solve
the problem. Also, this shell allows object-oriented programming using RULES and FACTS, RULES and OBJECTS to
solve the forecasting problem.
C. Solution Strategy
The procedures of the solution strategy is to ascertain the
accuracy, suitability, and credibility of selecting forecasting
method using the production rules in the rule base. An outline
of the solution strategy of the ES is shown in Fig. 6.
Based on the condition, input by the user, of the forecasting
objective and the situation of the power system to be used, the
engine emulates the initial state and the different system information. Then it checks the available forecasting methods to
match the forecasting objective. Afterwards, the procedure initiates the load forecasting computation with the most appropriated forecasting algorithms using the integrated system database.

KANDIL et al.: LONG-TERM LOAD FORECASTING FOR FAST DEVELOPING UTILITY

495

Fig. 8. Time series method for forecasting the normal developing system.

Fig. 7. Portion of main forecasting dialog of the proposed ES applied to

Egyptian system.

The engine takes the results of the forecasted peak load and
the produced errors to match the rule base of the forecasting criteria. If the output results satisfy the imposed constraints without
violations, the knowledge-based system terminates the analysis
process successfully. When the inference engine finds that the
imposed constraints such as selection of forecasting method are
not satisfied, it has to satisfy the goal using another forecasting
method that satisfies the corresponding rules. Therefore, the engine searches for the rules that match the pattern of describing
the forecast method. When the interfacing is completed after
generating the derived fact using forward chaining, the resulting
goal facts are displayed to user and stored in a file for possible
users request.

Fig. 9. Actual and forecasted load of the normal developing system

(19812001).

D. User Interface
The object of this module is to make the developed ES
friendly with its users. This object is achieved by designing
sets of Menus and Windows to help the user in his navigation
through the ES.

Fig. 10. Sector wise energy consumption models of the fast developing
system.

VI. IMPLEMENTATION OF THE DEVELOPED EXPERT SYSTEM

To demonstrate the effectiveness of the developed ES, two
cases are examined to produce the peak load forecasts for the
two systems explained above in this paper. The designed modules of the proposed ES are executed using Clips Shell, Visual
Basic 5, and spreadsheet packages. All modules have been executed on a PC-Pentium II.
Case 1) Normal Developing Utility: The proposed ES emulates time series method to perform the demand
forecast for the normal developing system as shown
in the Forecasting dialog of Fig. 7. Due to the stability of the system behavior the ES emulates the
polynomial trend model as shown in Fig. 8. in the
Forecasting dialog of Fig. 7. Due to the stability of
the system behavior the ES emulates the polynomial
trend model as shown in Fig. 8. This model was previously recommended by the classical method (see
Fig. 1). Using this model, the actual and forecasted
load from 1981 to 2008 is developed as shown in
Fig. 9.

Fig. 11. Actual and forecasted peak load of the fast developing system from
1985 to 2008.

Case 2) Fast Developing Utility: Employing the similar

process of case 1 using the fast developing system
information, simulation method is addressed as a
more reliable one. The best models of the energy
consumption are produced as shown in Fig. 10.
The best model for each category is developed and the
summation of the energy demand of all sectors forms the total
system energy. The forecasted peak load over the same period
is then produced as shown in Fig. 11 using the forecasted
energy, losses and the forecasted load factor.

496

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 17, NO. 2, MAY 2002

oped expert system can serve as a valuable assistant to system

planners in performing their annual load forecasting duties. Finally, it can be expected to serve as a valuable assistant also for
training purposes.
REFERENCES

Fig. 12. Energy demand and load forecast of each area for transmission
planning.
TABLE III
STATISTICAL ERRORS COMPARISON OF DIFFERENT FORECASTING
METHODS (19941998)

After the ES developing the forecast on the total system

basis, which is sufficient for generation planning, it produces
the load forecasting of each area, which is essentially needed
for transmission planning studies [10]. Thus, the ES through its
inference engine, reads the future load results and the heuristic
knowledge stored in its knowledge base to develop the peak
demand of each area. A summary of the relevant data of the
forecasted peak load of each area for 2001 appears in the
printout of Fig. 12.
VII. FORECASTING ACCURACY
The statistical error comparisons for all methods for the fast
developing system from 1994 to 1998 are shown in Table III.
This table indicates that the implementation of the long-term
forecasting using the ES is very close to actual recorded values
and the proposed ES is more appropriated to yield the best
system forecast.
VIII. CONCLUSIONS
The results of the classical forecasting methods concludes
that there is no single method, when applied to fast developing
system, has shown a constant and stable performance over the
forecasting period where as most of these models are valid when
applied to the normal utilities. In this paper, a long-term load
forecasting is developed and classified according to the forecasting problem using a knowledge-based expert system (ES).
The proposed methodology is applied successfully to forecast
yearly peak load for normal and fast developing power systems.
The results are found to be very close to the actual loads and to
be more accurate than those obtained from direct application.
Since the expert system is very flexible in updating the forecasting methods and heuristic rules, it is expected that the devel-

[1] H. L. Willis, Spatial Electric Load Forecasting. New York: Marcel

Dekker, 1996.
[2] D. W. Bunn and E. D. Farmer, Comparative Models for Electrical Load
Forecasting. New York: Wiley, 1985.
[3] N. E. Hasanien, An accurate model of weather sensitive for long-term
load forecasting for fast growing power system, in Proc. Second Regional Conf. CIGRE Committees Arab Countries, Amman, Jordan, May
1997.
[4] E. H. Barakat and M. A. M. Eissa, Forecasting monthly peak demand in
fast growing electric utility using a composite multi-regression decomposition model, Proc. Inst. Elect. Eng. C, vol. 136, no. 1, pp. 3541,
Jan. 1989.
[5] B. V. Stephen, Statistics for Engineering Problem Solving. Boston,
MA: PWS, 1994.
[6] Energy and Electricity Demand Forecasting for Nuclear Power Planning in Developing Countries. Vienna: IAEA, 1988.
[7] A. S. Al Fuhaid, M. A. El-Sayed, and M. S. Mahmoud, Cascaded artificial neural networks for short-term load forecasting, IEEE Trans.
Power Syst., vol. 12, pp. 15241529, Nov. 1997.
[8] J. C. Giarratano, CLIPS Expert Systems: NASA Software Technology
Branch, 1993, vol. I & II.
[9] M. S. Kandil, S. M. El-Debeiky, and N. E. Hasanien, Rule-based
system for determining units locations of a developed generation
expansion plan for transmission planning, Proc. Inst. Elect. Eng. Gen.
Transmiss., Distribut., vol. 147, no. 1, pp. 6268, Jan. 2000.
[10] H. L. Wills and H. N. Tram, Load forecasting for transmission planning, IEEE Trans. Power Syst., vol. 103, pp. 561568, Mar. 1984.

M. S. Kandil was born in Damanhour, Egypt, in 1938. He received the Ph.D.

degree from the University of Manchester Institute of Science and Technology,
Manchester, U.K., in 1967.
He was Professor, Head of the Electrical Department, and Dean of the Faculty
of Engineering of Mansoura University, Mansoura, Egypy, where he is now a
Professor. His technical activities include power system protection, operation,
security, planning, load forecast, and solar energy, as well as applications of
artificial intelligence to power system problems. He is a consultant to a number
of sub-utilities in Egypt.

S. M. El-Debeiky (SM77) was born in Egypt, in

1939. He received the Ph.D. degree from Cairo
University, Cairo, Egypt, in 1970. He was Head of
Electrical Department of the Faculty of Engineering
of Ain Shams University, Egypt, where he is
now a Professor. His technical activities include
high-voltage engineering, over-voltage studies,
and power system planning and operation. He is a
consultant for the Egyptian Power System Company.
He has written more than 75 papers in international
journals.
Dr. El-Debeiky is a member of IEE and CIGRE.

N. E. Hasanien was born in Dyarb Negm, Egypt, in

1957. He received the Ph.D. degree from Ain Shams
University in 2001.
He was with the Egyptian Electricity Authority
as a system studies and planning engineer since
1982. He has also been with the Saudi Electricity
Company, Jeddah, Saudi Arabia, since 1989, where
he is now a system planner. His technical activities include system analysis, system generation,
transmission planning, load forecast, switching and
over-voltage studies, and applications of artificial
intelligence to power system problems.