Abstract- Load forecasting
is the most important factor for any electric industry in the deregulated
economy structure. It has wide range of applications namely as energy
purchasing and generation, load switching, contract evaluation, and
infrastructure development. Numerous techniques are being used to get an
accurate forecasting. In this paper we are going to analyse some of them. We
have tried to understand the impact of load forecasting on the power system and
its limitations.
Keywords- Load,
forecasting, statistics, power system, artificial intelligence
I. Introduction
For
any utility company along with a good quality power that is mostly important is
a proper ‘load forecasting. It basically helps them to make important decisions
regarding decisions on purchasing and generation of electric power,
infrastructure development, and load switching/pooling. Importantly, it is
extremely important for energy suppliers, financial institutions, and other
participants in electric energy generation, transmission, distribution players,
and market condition.
Load
forecasts can be further characterised into three different categories:
short-term forecasting which usually varies from one hour to one week, medium-term
forecasting which usually varies from a week to a year, and the long-term
forecasting which are designed by considering more than a year time-period.
Moreover, the load forecasting for different time horizons are also very important
for different operations within a utility organisation. The basic nature of
these forecasting are very different as well. For example, for a particular
area, it may be possible to predict the very next day load with an accuracy of approximately
1-3%. However, it is just impossible to predict the next year peak power load
with the same magnitude of accuracy since accurate long-term weather forecasts
are not available till now. For the next year peak load forecast, it might be
possible to provide the probability distribution of the load on the basis of
historical weather observations.
Load
forecasting has always been a very important factor for taking a planning and
operational decision by a utility company. However, with the deregulation of
the energy industries, it has become a very significant factor. With the
variation in demand and supply, frequent changing of weather conditions and
increasing of energy prices by a factor
of ten or more during peak-load situations, it has become a vitally important
consideration factor for utility organisation. Short-term load forecasting helps
us to estimate the magnitude of load and to make a futuristic decision that can
prevent overloading. Timely implementation of such decision results into the
improvement of network reliability and reduction of occurrences of equipment
failures and blackouts. Moreover, Load forecasting is also very important
factor for contract evaluations and evaluations of various sophisticated financial
products on energy pricing being offered and supplied by the market forces. Decisions
on capital expenditures based on long-term forecasting are also an important
consideration factor in a non-deregulated economy when increase in rate could
be justified by the capital expenditure incurred on it.
Most forecasting methods use different
statistical techniques or artificial intelligence algorithms such as neural
networks, regression, fuzzy logic, etc. For the said purpose. For medium- and
long-term forecasting, generally two of the methods; so-called end-use and
econometric approach are used. As we know, a large variety of mathematical
tools and ideas are being extensively implemented for load forecasting. But
with the further development and improvements of appropriate mathematical tools
will lead to the development of a more precise, absolute and accurate load
forecasting techniques. Moreover, the accuracy of load forecasting does not
only depends on the load forecasting techniques, but also depends on the
forecasted weather scenario accuracy.
II. Important Factors being considered for load forecasting
For short-term
load forecasting several factors, such as time factors, weather data, and
possible target customers’ classes should be considered predominantly. The
medium- and long-term forecasting takes into account of a wide range of factors
namely, the historical load and weather data, the number of targeted customers in different categories,
the electrical appliances being used in the area and their characteristics
including age, the demographic and economic data and their forecasts, etc.
The
time factors taken into consideration include the time of the year, the day of
the week, and the hour of the day. Most important factor that we have to take
in consideration is that, there are important differences in load between
weekends and weekday. The load on different weekdays can also sometimes behave
very differently. For example, Fridays and Mondays as being adjacent to weekends,
may have structurally a very different loads in comparison to Tuesday through Thursday.
This is particularly very true during the summer time. Moreover, Holidays are
more difficult to be forecasted than non-holidays because of their relative unfrequented
occurrences of event.
Weather
conditions also influence the power-load. In fact, forecasted weather parameters
are the most important factors in short-term load forecasts. Various weather
variables such as temperature and humidity could also be considered for load
forecasting.
Most electric utilities provide service
to customers of various types such as residential, commercial, and industrial.
The electric-load pattern is significantly different for customers that belong
to different classes but is somewhat alike for customers within a class.
Therefore, most utility company distinguishes load behaviour on a
class-to-class basis [1].
III. Forecasting Methods
Over the last few decades a number of
forecasting methods have came into picture or been developed. Two of the
methods, so-called end-use and econometric approach are broadly used for
medium- and long-term forecasting. A variety of methods, which include the
so-called similar day approach, various regression models, neural networks,
time series expert systems, statistical learning algorithms and fuzzy logic,
are used for short-term load forecasting. . But with the further development and
improvements of appropriate mathematical tools will lead to the development of
a more precise, absolute and accurate load forecasting techniques. Statistical
techniques usually require a more advanced mathematical model that represents load
as a function of different factors namely; time, weather, and the targeted customer
class. The two main categories of such mathematical models are: multiplicative
models and additive models. It only differs with each other as whether the
forecast load is the product (multiplicative) of a number of factors or is the sum (additive) of a number of
components. For example, Chen et al. [2] presented an additive model that takes
the form of load prediction as the function of four components:
L =
Ln + Lw + Ls + Lr,
where L is the
total load, Ln represents the “normal” load, Lw represents the weather
sensitive part of the load, Ls is a special event component that create a
substantial deviation from the usual pattern of load, and Lr is a completely
random term, may be any harmonic.
Chen
et al. [2] also suggested electricity pricing as an additional term that can be
included in the said model. Naturally, electricity consumptions are also
affected by price decrease/increase. Large cost sensitive industrial and institutional
loads can have a very greater impact on loads.
A multiplicative model may be of the
form
L =
Ln • Fw • Fs • Fr
where
Ln is the normal load and the correction factors Fw, Fs, and Fr are some
positive integers that can increase or decrease the overall power-load. These
corrections are further entirely based on weather factor (Fw), special events
factor (Fs), and the random fluctuation factor (Fr). Factors such as load
growth (Fg) and electricity pricing (Fp) can also be included into it [3].
III.I. Medium and long term load forecasting methods
The econometric
modelling, end-use modelling, and their combinations are often mostly used
methods for medium- and long-term load forecasting. Descriptions of electrical appliances
used by customers, the sizes of their houses, ageing of equipment, technology
advancement, customer behaviour, and dynamics of population in a particular
region are usually included in the statistical and simulation models entirely
based on the so-called end-use approach. In addition, predominant economic
factors such as employment levels, per capita incomes, and electricity prices
are included or being considered significantly in econometric models. Most
often these models are used in combination with the end-use approach model.
Long-term
forecasting mainly includes the forecasting based on the population changes,
economic development, commercial construction, and the technology advancement. The
end-use approach used for it directly estimates energy consumption over a
period of time by using extensive information on end use and end users, such as
number of electrical appliances, the frequency of their use, its age, sizes of
houses, and so on.
Statistical
information along with dynamics of change of consumption by customers is the
basis for these types of forecasting.
Moreover,
End-use models focus mainly on the diverse uses of electricity in the
residential, commercial, and industrial sector. It is entirely based on the principle
that electricity demand is mainly because of customer’s demand for lightening,
heating, cooling, refrigeration, etc. Thus the end-use model primarily explains
‘energy demand’ as a function of the number of electrical appliances available
in the market [4].
Ideally this approach is very absolute
as well as accurate. However, it is very sensitive to the magnitude and quality
of end-use data provided to it. For example, in this method equipment age
distribution factor is also a very important factor for particular types of
electrical appliances. End-use forecasting need less historical data but
require more information regarding customers and their electrical-equipment.
IV.Econometric models
The
econometric technique approach model combines economic theory along with statistical
techniques for forecasting of electricity demand. The approach basically
predicts the relationship between energy consumption factors (dependent variables)
and the consumption influencing factors. The said relationships are predicted
by the method called as the least-squares method or time series methods.
One
of the options in these type of framework is to simply aggregate the finding
done by econometric approach, when consumption in different sectors (residential,
commercial, industrial, etc.) is calculated as the function of weather,
economic and other variable factors, and then prediction are being assembled
using recent historical data.
Moreover, Integration of the econometric
approach with the end-use approach introduces a much required behavioural
components into the end-use equations.
V.Statistical model-based learning Approach
The
end-use models and the econometric model approach basically require a large
magnitude of information regarding electrical-appliances, customer’s behaviour,
economics, etc. Therefore, their
application is quite complicated and requires a much human participation. In
addition such data is also often not easily available regarding a particular
set of customers and their utility behaviour. Moreover, this problem arises more
significantly if the utility organisation wants to conduct a next-year
forecasting for a sub-area, which are often referred as load pockets. In such
case, the amount of the work (should be performed) increases in a direct
proportion with the frequency of load pockets. In addition, the finding by
end-use approach model and econometric approach model for different load
pockets are also very different from each-other.
In order to reduce the complexions of
the medium-term forecasting, and making them more accurate and absolute, a new
model is developed by Feinberg et al.([5], [6]). It is basically a statistical
model that learns the different parameters of load model from the historical
data.
L(t)
= F(d(t), h(t)) • f(w(t)) + R(t)
where L(t) is
the real or actual load at a particular time t, d(t) is the days of the week,
h(t) is the hours of the day, F(d, h) is the daily and hourly component factor,
w(t) is the weather data factor that also includes the temperature and
humidity, f(w) is referred as the weather factor, and R(t) is any random error.
In fact, w(t) is a vector quantity that mainly
consists of the current as well as lagged weather variables. It reflects the
fact that electric load does not only depends on the current weather conditions
but also on the weather behaviour of previous hours and days. It is because of
the fact that in summer season the use of air conditioners increases when the
hot weather continues for several days.

Figure 12.1 presents an example of a scatter plot that compares the model and real parameters.

Figure 12.2 demonstrates the convergence of the correlation between the actual load and the model for the iteration process.
VI.Importance of load forecasting in power system
Load forecasting in power system is
basically a centralised and integral process in the operation as well as
planning of any electrical utility organisation. it generally includes the
absolute and accurate estimation of both the magnitudes and the geographical
location of electrical load over a different period of time of the planning
horizon. This is the reason because of which the importance of load forecasting
is increasing significantly in a typical power system. Some of the important of
them are:-
·
Helps to control the operation:
- it helps to make the system more efficient by optimizing it. System should be
operated only when it is required, and should produce the power in accordance
with the demand only. This method further also reduces the wastage of different
energy resources elements, which were being used without of getting any output.
·
Helps in planning: - with the
help of load forecasting finding, utility company can plan their operation
according to the perceived demand. If the prediction of demand is more than
present, they can look of establishing a new unit or may look for arranging
them from somewhere else to compensate the demand.
·
Helps in determination of
Tariff: - in a current scenario, most utility company determine their tariff
rate after the consideration of finding of load forecasting. Determination on
the basis of sales forecast helps them to earn more profit or to increase the
revenue.
·
Helps central grid to allocate
power to different sectors judicially: - in the absence of a proper accurate
load forecasting, there may be a scenario in which a sector is getting more
power than adequate and there is a complete black-out in the other sector. This
condition is not judicial. With the help of a proper load forecasting finding,
the power can be allocated to different sectors judicially.
·
Limitation of pollution to an
extent: - with the help of optimization of power system and a proper load
forecasting finding, we can also help the environment by limiting to get it
polluted up to a certain extent. This is because of the fact that we are
producing that much power only, which is actually required. Surplus power
production does not suffice any purpose neither to us nor to the environment;
rather pollute it significantly.
·
Helps in taking a decision
regarding dispatch of amount of power, unit commitment, fuel allocation to the
unit and off-line network analysis.
VII. Limitations associated with the load forecasting
·
Whether factor: - no one really
knows, what it will be in next few hours. This is the main reason because of
which no forecasting in which ‘whether’ plays any role into it, is that much
accurate.
·
Customer Behaviour: - customer
behaviour also changes frequently and it is quite a Hercules task to predict
the behaviour of a very large set of customers.
·
Government policies and
Regulations: - The policies and regulations set by the government also cause a
significant impact on the accuracy of load forecasting finding. For example: if
the government make it mandatory to use only LED light for their public;
naturally load occurring will be less as was forecasted earlier.
·
Spontaneity of situation: - it
also plays an important role. For example: in a power plant suddenly a grid
failed so that the load of that grid have to be shifted to other grids, and
this will result into high perceiving of load than expected.
VIII. Future Scope
In
this paper we have discussed about several statistical and artificial
intelligence approaches that have been implemented for short-term, medium-term,
and long-term electric load forecasting. Several statistical models and
algorithms that have been developed and implemented are quite structurally
complex. The accuracy of the forecasting is also one of the area that could be
further improved and developed
Researchers
should also try to simplify the boundaries of applicability of the developed
models and algorithms. So far, there is not a single model or algorithm which
is full-proof. The main reason for that is, utility service vary significantly
for diverse range of industrial, commercial, and residential customers. In
addition to that, it also varies in geographic, climatologic, economic, and
social characteristics.
Selecting
and implementing the most compatible algorithm by a utility organisation can be
done by testing the algorithms on real collected data of the customers.
As sated above, weather plays an
important role to influence the load forecasting. The conventional approach to short-term
load forecasting considers the forecasted weather scenario as an input. However,
most important recently developed technique in weather forecasting is referred
as ’so-called ensemble Approach’ which advocates of computing multiple
forecasts as an input. After that probability weights can also be assigned to
these ensembles. The main advantages of having a load forecasting in the
probabilistic form is that:-
·
it can lead to a more accurate hourly
forecast
·
The probabilistic description of the
futuristic load can also be used as a significant finding to decision support
systems to make important decision regarding generation, purchasing, and
switching.
Conclusion
In a current scenario of a de-regulated
market structure ‘Accurate load forecasting’ has become a very important factor
for electric utility company in order to remain a significant player in the
market. In this paper we analysed some statistical and artificial intelligence
approaches that are being frequently used for electric load forecasting. We
also discussed some of factors that affect the accuracy of the finding of
forecast predominantly namely, as weather spontaneity, time factors, customer
behaviour, as well as economical and end use factors.
Recently, advanced mathematical modelling is also being used for the said purpose. Further improvement in the accuracy of load forecasting can also be implemented by providing short-term electrical load forecasting in the form of probability distributions rather than in the forecasted numbers; for example: the’ so-called ensemble approach’ can also be used. We have also seen the significant impact of load forecasting on the performance of power system. It basically helps to optimize the power system in accordance with the demand of customer.
Recently, advanced mathematical modelling is also being used for the said purpose. Further improvement in the accuracy of load forecasting can also be implemented by providing short-term electrical load forecasting in the form of probability distributions rather than in the forecasted numbers; for example: the’ so-called ensemble approach’ can also be used. We have also seen the significant impact of load forecasting on the performance of power system. It basically helps to optimize the power system in accordance with the demand of customer.
References
[1]
H.L. Willis.
Spatial Electric Load Forecasting. Marcel Dekker, New York, 1996.
[2]
H. Chen, C.A. Canizares, and A. Singh.
ANN-Based Short-Term Load Forecasting in Electricity Markets. Proceedings of
the IEEE Power Engineering Society Transmission and Distribution
Conference, 2:411–415, 2001.
[3]
S. Rahman. Formulation and Analysis of a
Rule-Based Short-Term Load Forecasting Algorithm. Proceedings of the IEEE,
78:805–816, 1990.
[4]
C.W. Gellings. Demand Forecasting for
Electric Utilities. The Fairmont Press, Lilburn, GA, 1996.
[5]
E.A. Feinberg, J.T. Hajagos, and D.
Genethliou. Load Pocket Modeling. Proceedings of the 2nd IASTED
International Conference:Power and Energy Systems, 50–54, Crete, 2002.
[6]
E.A. Feinberg, J.T. Hajagos, and D.
Genethliou. Statistical Load Modeling. Proceedings of the 7th IASTED International
Multi-Conference: Power and Energy Systems, 88–91, Palm Springs, CA, 2003.
[7]
J.W. Taylor and R. Buizza. Using Weather
Ensemble Predictions in Electricity Demand Forecasting. International
Journal of Forecasting, 19:57–70, 2003.
J.W.
Taylor and R. Buizza. Neural Network Load Forecasting With Weather Ensemble
Predictions. IEEE Transactions on Power Systems, 17:626–632, 200
Comments
Post a Comment