518 Ass 2
518 Ass 2
518 Ass 2
Abstract
This paper picturesquely depicts the comparison of different methodologies adopted for
predicting the load demand and highlights the changing trend and values under new
circumstances using latest non analytical soft computing techniques employed in the field of
electrical load forecasting. A very clear advocacy about the changing trends from conventional
and obsolete to the modern techniques is explained in very simple way. Load forecast has been
a central and an integral process in the planning and operation of electric utilities. Many
techniques and approaches have been investigated to tackle this problem in the last two
decades. These are often different in nature and apply different engineering considerations and
economic analysis. Further a clear comparison is also presented between the past standard
practices with the current methodology of electrical load demand forecasting. Besides all this,
different important points are highlighted which need special attention while doing load
forecasting when the environment is competitive and deregulated one.
1.0 INTRODUCTION
Electrical Load Forecasting is the estimation for future load by an industry or utility company.
on
Load forecasting is vitally important for the electric industry in the deregulated economy. A large
variety of mathematical methods have been developed for load forecasting. It has many
si
applications including energy purchasing and generation, load switching, contract evaluation,
and infrastructure development.
r
ve
Now a days, development in every sector is a heading at a very rapid pace and in the same
pattern, the demand for power is also growing. While speaking about electrical power, it is
al
important to understand that it has three main sectors i.e. generation, transmission and
distribution. Electrical power generated by any source is then transmitted through transmission
tri
lines at different voltage level and then distributed to different categories of consumers later on.
It is not as simple as described in few words but every stage is a complete independent system
ne
in itself. Effective load forecasts can help to improve and properly plan these three fields of
power systems [1].
hi
Accurate models for electric power load forecasting are essential to the operation and planning
ac
of a utility company. Load forecasting helps an electric utility to make important decisions
including decisions on purchasing and generating electric power, load switching, and
fM
infrastructure development. Load forecasts are extremely important for energy suppliers, ISOs,
financial institutions, and other participants in electric energy generation, transmission,
pd
Over the past decade, many western nations have begun major structural reforms of their
electricity markets. These reforms are aimed at breaking up traditional regional monopolies and
replacing them with several generation and distribution utilities that bid to sell or buy electricity
through a wholesale market. While the rules of how various wholesale markets operate differ, in
each case it is hoped that the end result is a decline in the price of electricity to end users and a
price that better reflects the actual costs involved. To successfully operate in these new markets
electricity utilities face two complex statistical problems: how to forecast both electricity load and
the wholesale spot price of electricity. Failure to implement efficient solutions to these two
forecasting problems can directly result in multimillion dollar losses through uninformed trades in
the wholesale market.
_____________________________________________________________________________________________
Electric Load
Forecasts
commitment and maximizing the utility revenues in the deregulated environment. Medium term
ve
electric load forecast spans the period from one week to several weeks, it is mainly utilized for
predicting the necessary power to purchase or sell from other neighboring networks (inter-tie
al
exchanged power) and also the fuel required by the utility in the near future. In short and
medium electric load forecast, it is required to know how much power we will need and at what
tri
time of the day; the information regarding where this demand is required is not of a major
concern [1].
ne
method uses data on electricity prices, income, population, the economy, and the growth rates
for each and then varies the mix according to varying sets of assumptions. Different
fM
assumptions produce different outcomes. The relationships between electricity demand and the
multitude of factors that influence or affect electricity demand are expressed in mathematical
pd
Energy planners often speak of scenarios. Hypothetical pictures of the future based on different
assumptions about economic or political events. They make different projections for each
scenario. For example, a low growth scenario might assume high energy prices and slow
population growth, while a high-growth scenario would assume the opposite. These scenarios
allow planners to see how electricity demand might change if the different assumed economic
and political events actually occur. All of the forecasting methods are capable of looking at
29
different scenarios and do so by changing their basic assumptions [5].
High forecasting accuracy and speed are the two most important requirements of short-term
load forecasting and it is important to analyze the load characteristics and identify the main
factors affecting the load. In electricity markets, the traditional load affecting factors such as
season, day type and weather, electricity price that have voluntary and may have a complicated
relationship with system load..
Various forecasting techniques have been applied to short-term load forecasting to improve
accuracy and efficiency. In general, these techniques can be classified as either traditional or
modern. Traditional statistical load forecasting techniques, such as regression, time series,
pattern recognition, Kalman filters, etc., have been used in practice for a long time, showing the
forecasting accuracy that is system dependent. These traditional methods can be combined
using weighted multi-model forecasting techniques, showing adequate results in practical
systems. However, these methods cannot properly represent the complex nonlinear
relationships that exist between the load and a series of factors that influence it, which are
typically dependent on system changes (e.g., season or time of day).
on
The short term load forecasting methods are:
si
r
Fuzzy logic
Support Vector Machines
ne
Similar day approach is based on searching historical data of days of one, two or three years
ac
having the similar characteristics to the day of forecast. The characteristics include similar
weather conditions, similar day of the week or date. The load of the similar day is considered as
the forecast. Now, instead of taking a single similar day, forecasting is done through linear
fM
combinations or regression procedures by taking several similar days. The trend coefficients of
the previous years are extracted from the similar days and forecast of the concern day is done
pd
on their basis.
Regression is the one of most widely used statistical techniques. For electric load forecasting,
regression methods are usually used to model the relationship of load consumption and other
factors such as weather, day type, and customer class. There are several regression models for
the next day peak forecasting. Their models contain deterministic influences such as holidays,
random variables influences such as average loads, and exogenous influences such as
weather.
Time series models can be accurate in some situations, but are especially complex and require
large amounts of historical data. Additionally, careful efforts must made to ensure an accurate
time line through out data collection filtering modeling and recall processes. Time series
analysis widely used in the martial management for forecasting of customer demand for goods
services. Time series approaches are not widely used for energy industry forecasting. Because
they typically do not take into account other key factor, such as weather forecasts [3].
on
Time series have been used for longtime in such fields as economics, digital signal processing,
si
as well as electric load forecasting. In particular, ARMA (autoregressive moving average),
r
with exogenous variables), and ARIMAX (autoregressive integrated moving average with
exogenous variables) are the most used classical time series methods.
al
ARMA models are usually used for stationary processes while ARIMA is an extension of ARMA
tri
for non-stationary processes. ARMA and ARIMA use the time and load as the only input
parameters. Since load generally depends on the weather and time of the day, ARIMAX is the
ne
most natural tool for load forecasting among the classical time series models.
hi
structure of the brain. We know that the brain basically learns from the experience. The
biological inspired methods are thought to be the major advancement in the computational
fM
industry. In a neural network, the basic processing element is the neuron. These neurons get
input from some source, combine them, perform all necessary operations and put the final
pd
results on the output. Artificial neural networks are developed since mid-1980 and extensively
applied. They have very successful applications in pattern recognition and many other
problems.
Forecasting is based on the pattern observed from the past event and estimates the values for
the future. ANN is well suited to forecasting for two reasons. First, it has been demonstrated that
ANN are able to approximate numerically any continuous function to be desired accuracy. In
this case the ANN is seen as multivariate, nonlinear and nonparametric methods. Secondly,
ANNs are date-driven methods, in the sense that it is not necessary for the researcher to use
tentative modals and then estimate their parameters. ANNs are able to automatically map the
relationship between input and output, they learn this relationship and store this learning into
their parameters [3].
The first way is by repeatedly forecasting one hourly load at a time. The second way is by using
31
a system with 24 NNs in parallel, one for each hour of the day. Estimating a model that fits the
data so well that it ends by including some of In Multi Layer Perceptron(MLP) structure of neural
network, the most commonly training algorithm use is the back propagation algorithm. These
algorithms are iterative; some criteria must be defined to stop the iterations. For this either
training is stopped after a fixed number of iterations or after the error decreased below some
specified tolerance. This criterion is not adequate, this insure that the model fits closely to the
training data but does not guarantee of good performance they may lead to over-fitting of the
model. "Over-fitting" means the error randomness in its structure, and then produces poor
forecasts. MLPs model is over-trained or because it is too complex. One way to avoid
overtraining is by using cross-validation. The sample set is split into a training set and a
validation set. The neural network parameters are estimated on the training set, and the
performance of the model is tested, every few iterations, on the validation set. When this
performance starts to deteriorate (which means the neural network is over-fitting the training
data), the iterations are stopped, and the last set of parameters to be computed is used to
produce the forecasts. Nowadays, other than MLPs to avoid the problems of over-fitting and
over-parameterization, the ANNs architectures used for prediction of electrical load are
Functional Link Network (FLN) model [1].
To use the ANN in electric load forecast problems, distribution engineers should decide upon a
number of basic variables, these variables include:
Activation functions
ve
Expert systems are new techniques that have come out as a result of advances in the field of
artificial intelligence (AI) in the last two decades. An expert system is a computer program,
ne
which has the ability to act as an expert. This means this computer program can reason,
explain, and have its knowledge base expanded as new information becomes available to it.
hi
ac
The load forecast model is built using the knowledge about the load forecast domain from an
expert in the field. The "Knowledge Engineer" extracts this knowledge from load forecast
fM
(domain) expert which is called the acquisition module component of the expert system. This
knowledge is represented as facts and rules by using the first predicate logic to represent the
facts and IF-THEN production rules. This representation is built in what is called the knowledge
pd
base component of the expert system. The search for solution or reasoning about the
conclusion drawn by the expert system is performed by the "Inference Engine" component of
the expert system. For any expert system it has to have the capability to trace its reasoning if
asked by the user. This facility is built through an explanatory interface component.
An example demonstrating this approach is the rule-based algorithm which is based on the work
of two scientists Rahman and Baba. This algorithm consists of functions that have been
developed for the load forecast model based on the logical and syntactical relationship between
the weather and prevailing daily load shapes in the form of rules in a rule-base. The rule-base
developed consists of the set of relationships between the changes in the system load and
changes in natural and forced condition factors that affect the use of electricity. The extraction of
these rules was done off-line, and was dependent on the operator experience and observations
by the authors in most cases. Statistical packages were used to support or reject some of the
possible relationships that have been observed
32
The rule-base consisted of all rules taking the IF-THEN form and mathematical expressions.
This rule-base is used daily to generate the forecasts. Some of the rules do not change over
time, some change very slowly while others change continuously and hence are to be updated
from time to time [4].
The advantage of fuzzy logic is that there is no need of mathematical models for mapping
between inputs and outputs and also there is no need of precise or even noise free inputs.
Based on the general rules, properly designed fuzzy logic systems are very strong for the
electrical load forecasting. There are many situations where we require the precise outputs.
After the whole processing is done using the fuzzy logic, the “defuzzification” is done to get the
precise outputs.
We know that power system load is influenced by many load factors such weather, economic
and social activities and different load components. By the analysis of historical load data it is
not easy to make the accurate forecast. The use of these intelligent methods like fuzzy logic and
expert systems provide advantage on other conventional methods. The numerical aspects and
on
uncertainties are suitable for the fuzzy methodologies[5].
si
4.7 Support Vector Machines
r
Support Vector Machines (SVM) are the most powerful and very recent techniques for the
ve
solution of classification and regression problems. This approach was come to known from the
work of Vapnik’s, his statistical learning theory. Other from the neural network and other
al
intelligent systems, which try to define the complex functions of the inputs, support vector
machines use the nonlinear mapping of the data in to high dimensional features by using the
tri
kernel functions mostly. In support vector machines, we use simple linear functions to create
linear decision boundaries in the new space. In the case of neural network, the problem is in the
ne
choosing of architecture and in the case of support vector machine, problems occurs in
choosing a suitable kernel.
hi
Mohandes applied a method of support vector machines for short-term electrical load
ac
forecasting. He compares its method performance with the autoregressive method. The results
indicate that SVMs compare favorably against the autoregressive method. Chen also proposed
fM
a SVM model to predict daily load demand of a month. Lots of methods are used in support
vector machines [3].
pd
1. Trend Analysis
2. End Use Analysis
3. Econometrics
Each of the three forecasting methods uses a different approach to determine electricity
demand during a specific year in a particular place. Each forecasting method is distinctive in its
handling of the four basic forecast ingredients:
33
1. The mathematical expressions of the relationship between electricity demand and the
factors which influence or affect it - the function
2. The factors which actually influence electricity demand (population, income, prices, etc.)
- the independent variables
3. Electricity demand itself - the dependent variable
4. How much electricity demand changes in response to population, income, price, etc.,
changes- the elasticities?
The only way to determine the accuracy of any load forecast is to wait until the forecast year
has ended and then compare the actual load to the forecast load. Even though the whole idea
of forecasts is accuracy, nothing was said in the comparison of the three forecasting methods
about which method produces the most accurate forecasts. The only thing certain shut any
long-range forecast is that it can never be absolutely precise. Forecasting accuracy depends on
the quality and quantity of the historical data used, the validity of the forecasters basic
assumptions, and the accuracy of the forecasts of the demand-influencing factors (population,
income, price, etc.). None of these is ever perfect. Consequently, regional load forecasts are
reviewed some are revised yearly. Even so, there is simply electricity demand will be exactly as
forecast, no is used or who makes the forecast. Continually, and no assurance that matter what
method is used or who makes the forecast [3].
their forecasts based on their knowledge of future developments which might make future
ve
when there is not enough data to use more sophisticated methods or when time and funding do
not allow for a more elaborate approach.
tri
The disadvantage of a trend forecast is that it produces only one result - future electricity
ne
demand. It does not help analyze why electricity demand behaves the way it does, and it
provides no means to accurately measure how changes in energy prices or government policies
hi
(for instance) influence electricity demand. Because the assumptions used to make the forecast
(informed judgments) are usually not spelled out, there is often no way to measure the impact of
ac
a change in one of the assumptions. Another shortcoming of trend analysis is that it relies on
past patterns of electricity demand to project future patterns of electricity demand. This
fM
simplified view of electrical energy could lead to inaccurate forecasts in times of change,
especially when new concepts such as conservation and load management must be included in
pd
The disadvantage of end-use analysis is that most end-use models assume a constant
relationship between electricity and end-use (electricity per appliance, or electricity used per
dollar of industrial output). This might hold true over a few years, but over a 10-or 20-year
period, energy savings technology or energy prices will undoubtedly change, and the
relationships will not remain constant. End-use analysis also requires extensive data, since all
relationships between electric load and all the many end-uses must be calculated as precisely
as possible. Data on the existing stock of energy-consuming capital (buildings, machinery, etc.)
in many cases is very limited. Also, if the data needed for end-use analysis is not current, it may
not accurately reflect either present or future conditions, and this can affect the accuracy of the
forecast. Finally, end-use analysis, without an econometric component that is explained above,
does not take price changes (elasticity of demand) in electricity or other competing fuels into
consideration.
Ideally this approach is very accurate. However, it is sensitive to the amount and quality of end-
on
use data. For example, in this method the distribution of equipment age is important for
particular types of appliances. End-use forecast requires less historical data but more
si
information about customers and their equipment [1].
r
ve
5.3 Econometric
Econometrics uses economics, mathematics, and statistics to forecast electricity demand.
al
Econometrics is a combination of trend analysis and end-use analysis, but it does not make the
trend-analyst’s assumption that future electricity demand can be projected based on past
tri
demand. Moreover, unlike many end-use models, econometrics can allow for variations in the
relationship between electricity input and end-use.
ne
electricity demand and the factors which influence that demand. For instance, an equation can
show how electricity demand in the past reacted to population growth, price changes, etc. For
ac
each influencing factor, the equation can show whether the factor caused an increase or
decrease in electricity demand, as well as the size (in percent) of the increase or decrease. For
fM
price changes, the equation can also show how long it took consumers to respond to the
changes. The equation is then tested and fine tuned to make sure that it is as reliable a
pd
representation as possible of the past relationships. Once this is done, projected values of
demand-influencing factors (population, income, prices) are put into the equation to make the
forecast. A similar procedure is followed for all of the equations in the model.
The advantages of econometrics are that it provides detailed information on future levels of
electricity demand, why future electricity demand increases or decreases, and how electricity
demand is affected by various factors. In addition, it provides separate load forecasts for
residential, commercial, and industrial sectors. Because the econometric model is defined in
terms of a multitude of factors (policy factors, price factors, end-use factors), it is flexible and
useful for analyzing load growth under different scenarios.
Also, the econometric load forecast can only be as accurate as the forecasts of factors which
influence demand. Because the future is not known, projections of very important demand-
influencing factors such as electricity, natural gas, or oil prices over a 10- or 20-year period are,
at best, educated guesses. Finally) many of the demand-influencing factors which may be
treated and projected individually in the mathematical equations could actually depend on each
other, and it is difficult to determine the nature of these interrelationships. For example, higher
industrial electricity rates may decrease industrial employment, and projecting both of them to
increase at the same time may be incorrect. A model which treats projected industrial electricity
rates and industrial employment separately would not show this fact.
Econometric models work best when forecasting at national, regional, or state levels. For
smaller geographical areas, meeting the model can be a problem. This is oddly shaped service
areas for which there demographic data.
The method for short-term forecasting are similar day approach, various regression models,
ve
time series, neural networks, statistical learning algorithms, fuzzy logic, and expert systems.
Similar day approach is based on searching historical data of days of one, two or three years
al
having the similar characteristics to the day of forecast. Regression is the one of most widely
used statistical techniques. For electric load forecasting, regression methods are usually used to
tri
model the relationship of load consumption and other factors such as weather, day type, and
customer class. There are several regression models for the next day peak forecasting. Their
ne
models contain deterministic influences such as holidays, random variables influences such as
average loads, and exogenous influences such as weather. Time series is a very popular
hi
approach for the electrical load forecasting. Two important models of time series are ARMA and
ARIMA. ARMA and ARIMA use the time and load as the only input parameters. Since load
ac
generally depends on the weather and time of the day, ARIMAX is the most natural tool for load
forecasting among the classical time series models [2].
fM
The other methods are based on Artificial intelligence, they are called Intelligent Systems. In
pd
Artificial Neural Network, forecasting is based on the pattern observed from the past event
and estimates the values for the future. ANN is well suited to forecasting for two reasons. First,
it has been demonstrated that ANN are able to approximate numerically any continuous function
to be desired accuracy. In this case the ANN is seen as multivariate, nonlinear and
nonparametric methods. Secondly, ANNs are date-driven methods, in the sense that it is not
necessary for the researcher to use tentative modals and then estimate their parameters. ANNs
are able to automatically map the relationship between input and output, they learn this
relationship and store this learning into their parameters. An Expert System is a computer
program, which has the ability to act as an expert. This means this computer program can
reason, explain, and have its knowledge base expanded as new information becomes available
to it. The load forecast model is built using the knowledge about the load forecast domain from
an expert in the field. This knowledge is represented as facts and rules by using the first
predicate logic to represent the facts and IF-THEN production rules. This representation is built
in what is called the knowledge base component of the expert system. The search for solution
or reasoning about the conclusion drawn by the expert system is performed by the "Inference
36
Engine" component of the expert system. For any expert system it has to have the capability to
trace its reasoning if asked by the user. This facility is built through an explanatory interface
component. Fuzzy logic based on the usual Boolean logic which is used for digital circuit
design. In case of fuzzy logic, the input is related to the comparison based on qualities. The
advantage of fuzzy logic is that there is no need of mathematical models for mapping between
inputs and outputs and also there is no need of precise or even noise free inputs. Based on the
general rules, properly designed fuzzy logic systems are very strong for the electrical load
forecasting.
The methods for long- and medium-term forecasting are trend analysis, end-use and
econometric approach. The advantage of trend analysis is that it is quick, simple and
inexpensive to perform and does not require much previous data. The basic idea of the end-use
analysis is that the demand for electricity depends what it use for (the end-use). The
advantages of end-use analysis is that it identifies exactly where electricity goes, how much is
used for each purpose, and the potential for additional conservation for each end-use. The
disadvantage of end-use analysis is that most end-use models assume a constant relationship
between electricity and end-use (electricity per appliance, or electricity used per dollar of
industrial output). This might hold true over a few years, but over a 10-or 20-year period, energy
savings technology or energy prices will undoubtedly change, and the relationships will not
remain constant. The advantages of econometrics are that it provides detailed information on
future levels of electricity demand, why future electricity demand increases or decreases, and
how electricity demand is affected by various factors. A disadvantage of econometric
forecasting is that in order for an econometric forecast to be accurate, the changes in electricity
demand caused by changes in the factors influencing that demand must remain the same in the
on
forecast period as in the past [5].
si
7.0 CONCLUSION
Modern load forecasting techniques, such as expert systems, Artificial Neural Networks (ANN),
r
ve
fuzzy logic, wavelets, have been developed recently, showing encouraging results. Among
them, ANN methods are particularly attractive, as they have the ability to handle the nonlinear
relationships between load and the factors affecting it directly from historical data.
al
The trend analysis, end-use modeling and econometric modeling are the most often used
tri
methods for medium- and long-term load forecasting. Trend analysis (trending) extends past
growth rates of electricity demand into the future, using techniques that range from hand-drawn
ne
behavior, and population dynamics are usually included in the statistical and simulation models
based on the so-called end-use approach. In addition, economic factors such as per capita
ac
incomes, employment levels, and electricity prices are included in econometric models. These
models are often used in combination with the end-use approach. Long-term forecasts include
fM
the forecasts on the population changes, economic development, industrial construction, and
technology development.
pd
REFERENCES
[1] “Computational Intelligence in Time Series Forecasting Theory and Engineering
Applications” (Advances in Industrial Control) by : Ajoy K. Palit, Dobrivoje Popovic,
Springer, 2005.
[2] Ibrahim Mogharm , Saifur Rehaman, “Analysis and Evaluation of Five Short Term Load
Forecasting Techniques”, IEEE Transactions on Power Systems, Vol. 4 No. 4, October
1989.
[3] H.L.Willis, “Distribution load forecasting”, IEEE Tutorial course on power distribution
planning, EHO 361-6-PWR, 1992.
[4] J.V. Ringwood “Intelligent Forecasting of Electricity Demand”.
www.forecastingprinciples.com
[5] Andrew P. Douglas, Arthur M. Breipohl, “Risk Due To Load Forecast Uncertainty in
Short Term Power System Planning” IEEE Transactions on Power Systems, Vol. 13 No.
4, November, 1998.
37