Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
denomination in the previous week. Secondly, future date. In projecting electricity demand, a
there are many important exogenous variables method uses data on electricity prices, income,
that must be considered, especially weather- population, the economy, and the growth rates
related variables. It is relatively easy to get for each and then varies the mix according to
forecast with about 10 % mean absolute error; varying sets of assumptions. Different
however, the cost of error are so high that assumptions produce different outcomes. The
research could help to reduce it in a few percent relationships between electricity demand and the
points would be amply justified [2]. multitude of factors that influence or affect
electricity demand are expressed in mathematical
equations called functions. A model is a
2.0 ELECTRICAL LOAD collection of functions. A function, in turn, is
FORECASTING TYPES made up of variables for which those factors
which change or can be changed. Independent
The electricity supply industry requires to variables are those factors which influence the
forecast electricity demand with lead times that demand for electricity, and the dependent
range from the short term (a few minutes, hours, variable is electricity demand itself. In other
or days ahead) to the long term (up to 20 years words, the demand for electricity depends on
ahead). Load forecasting has three techniques population, income, prices, etc. Finally,
shown in Figure 2.1: elasticities describe how much the dependent
variable (electricity demand) changes in sense to
small changes in the independent variables.
Electric Load Elasticities are what the modeler uses to measure
Forecasts
consumer behavior.
affecting the load. In electricity markets, the previous years are extracted from the similar
traditional load affecting factors such as season, days and forecast of the concern day is done on
day type and weather, electricity price that have their basis.
voluntary and may have a complicated
relationship with system load.. 4.2 Regression Based Approach
Various forecasting techniques have been The term "regression" was used in the nineteenth
applied to short-term load forecasting to improve century to describe a biological phenomenon,
accuracy and efficiency. In general, these namely that the progeny of exceptional
techniques can be classified as either traditional individuals tend on average to be less
or modern. Traditional statistical load forecasting exceptional than their parents and more like their
techniques, such as regression, time series, more distant ancestors.
pattern recognition, Kalman filters, etc., have
been used in practice for a long time, showing
Linear regression is a technique which examines
the forecasting accuracy that is system
the dependent variable to specified independent.
dependent. These traditional methods can be
The independent variables are firstly considered
combined using weighted multi-model
because changes occur in them unfortunately. In
forecasting techniques, showing adequate results
energy forecasting, the dependent variable is
in practical systems. However, these methods
usually demand or price of the electricity
cannot properly represent the complex nonlinear
because it depends on production which on the
relationships that exist between the load and a
other hand depends on the independent variables.
series of factors that influence it, which are
Independent variables are usually weather
typically dependent on system changes (e.g.,
related, such as temperature, humidity or wind
season or time of day).
speed. Slope coefficients measure the sensitivity
of the dependent variable that how they changes
The short term load forecasting methods are
with the independent variable. Also, by
measuring how significant each independent
Similar Day Lookup Approach variable has historically been in its relation to the
dependent variable. The future value of the
Regression Based Approach dependent variable can be estimated. Essentially,
regression analysis attempts to measure the
Time Series Analysis degree of correlation between the dependent and
independent variables, thereby establishing the
Artificial Neural Network latters predicted values[3].
historical data as input, time series analysis fits a from the past event and estimates the values for
model according to seasonality and trend. the future. ANN is well suited to forecasting for
two reasons. First, it has been demonstrated that
Time series models can be accurate in some ANN are able to approximate numerically any
situations, but are especially complex and require continuous function to be desired accuracy. In
large amounts of historical data. Additionally, this case the ANN is seen as multivariate,
careful efforts must made to ensure an accurate nonlinear and nonparametric methods. Secondly,
time line through out data collection filtering ANNs are date-driven methods, in the sense that
modeling and recall processes. Time series it is not necessary for the researcher to use
analysis widely used in the martial management tentative modals and then estimate their
for forecasting of customer demand for goods parameters. ANNs are able to automatically map
services. Time series approaches are not widely the relationship between input and output, they
used for energy industry forecasting. Because learn this relationship and store this learning into
they typically do not take into account other key their parameters [3].
factor, such as weather forecasts [3].
The first way is by repeatedly forecasting one
Time series have been used for longtime in such hourly load at a time. The second way is by
fields as economics, digital signal processing, as using a system with 24 NNs in parallel, one for
well as electric load forecasting. In particular, each hour of the day. Estimating a model that fits
ARMA (autoregressive moving average), the data so well that it ends by including some of
ARIMA (autoregressive integrated moving In Multi Layer Perceptron(MLP) structure of
average), ARMAX (autoregressive moving neural network, the most commonly training
average with exogenous variables), and algorithm use is the back propagation algorithm.
ARIMAX (autoregressive integrated moving These algorithms are iterative; some criteria
average with exogenous variables) are the most must be defined to stop the iterations. For this
used classical time series methods. either training is stopped after a fixed number of
iterations or after the error decreased below some
ARMA models are usually used for stationary specified tolerance. This criterion is not
processes while ARIMA is an extension of adequate, this insure that the model fits closely
ARMA for non-stationary processes. ARMA and to the training data but does not guarantee of
ARIMA use the time and load as the only input good performance they may lead to over-fitting
parameters. Since load generally depends on the of the model. "Over-fitting" means the error
weather and time of the day, ARIMAX is the randomness in its structure, and then produces
most natural tool for load forecasting among the poor forecasts. MLPs model is over-trained or
classical time series models. 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
4.4 Artificial Neural Networks validation set. The neural network parameters are
estimated on the training set, and the
Artificial Neural Networks are still at very early performance of the model is tested, every few
stage electronic models based on the neural iterations, on the validation set. When this
structure of the brain. We know that the brain performance starts to deteriorate (which means
basically learns from the experience. The the neural network is over-fitting the training
biological inspired methods are thought to be the data), the iterations are stopped, and the last set
major advancement in the computational of parameters to be computed is used to produce
industry. In a neural network, the basic the forecasts. Nowadays, other than MLPs to
processing element is the neuron. These neurons avoid the problems of over-fitting and over-
get input from some source, combine them, parameterization, the ANNs architectures used
perform all necessary operations and put the final for prediction of electrical load are Functional
results on the output. Artificial neural networks Link Network (FLN) model [1].
are developed since mid-1980 and extensively To use the ANN in electric load forecast
applied. They have very successful applications problems, distribution engineers should decide
in pattern recognition and many other problems. upon a number of basic variables, these variables
include:
Forecasting is based on the pattern observed
IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 5, No 3, September 2011
ISSN (Online): 1694-0814
www.IJCSI.org 508
Input variable to the ANN (load, facility is built through an explanatory interface
temperatureetc) component.
Trend analysis (trending) extends past growth and multiplying again by the projected number
rates of electricity demand into the future, using of homes, an estimate of how much electricity
techniques that range from hand-drawn straight will be needed to run all household appliances in
lines to complex computer-produced curves. a geographical area during any particular year in
These extensions constitute the forecast. Trend the future can be determined. Using similar
analysis focuses on past changes or movements techniques for electricity used in business and
in electricity demand and uses them to predict industry, and then adding up the totals for
future changes in electricity demand. Usually, residential, commercial, and industrial sectors, a
there is not much explanation of why demand total forecast of electricity demand can be
acts as it does, in the past or in the future. derived. The advantages of end-use analysis is
Trending is frequently modified by informed that it identifies exactly where electricity goes,
judgment, wherein utility forecasters modify how much is used for each purpose, and the
their forecasts based on their knowledge of potential for additional conservation for each
future developments which might make future end-use. End-use analysis provides specific
electricity demand behave differently than it has information on how energy requirements can be
in the past. reduced over time from conservation measures
such as improved insulation levels, increased use
The advantage of trend analysis is that it is of storm windows, building code changes, or
simple, quick and inexpensive to perform. It is improved appliance efficiencies. An end-use
useful when there is not enough data to use more model also breaks down electricity into
sophisticated methods or when time and funding residential, commercial and industrial demands.
do not allow for a more elaborate approach. Such a model can be used to forecast load
changes caused by changes within one sector
The disadvantage of a trend forecast is that it (residential, for example) and load changes
produces only one result - future electricity resulting indirectly from changes in the other
demand. It does not help analyze why electricity two sectors. Commercial sector end-use models
demand behaves the way it does, and it provides currently being developed have the capability of
no means to accurately measure how changes in making energy demand forecasts by end-uses as
energy prices or government policies (for specific as type of business and type of building.
instance) influence electricity demand. Because This is a major improvement over projecting
the assumptions used to make the forecast only sector-wide energy consumption and using
(informed judgments) are usually not spelled out, economic and demographic data for large
there is often no way to measure the impact of a geographical areas [1].
change in one of the assumptions. Another
shortcoming of trend analysis is that it relies on The disadvantage of end-use analysis is that
past patterns of electricity demand to project most end-use models assume a constant
future patterns of electricity demand. This relationship between electricity and end-use
simplified view of electrical energy could lead to (electricity per appliance, or electricity used per
inaccurate forecasts in times of change, dollar of industrial output). This might hold true
especially when new concepts such as over a few years, but over a 10-or 20-year
conservation and load management must be period, energy savings technology or energy
included in the analysis [3]. 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
5.2 End Use Analysis many end-uses must be calculated as precisely as
possible. Data on the existing stock of energy-
The basic idea of end-use analysis is that the consuming capital (buildings, machinery, etc.) in
demand for electricity depends on what it is used many cases is very limited. Also, if the data
for (the end-use). For instance, by studying needed for end-use analysis is not current, it may
historical data to find out how much electricity is not accurately reflect either present or future
used for individual electrical appliances in conditions, and this can affect the accuracy of the
homes, then multiplying that number by the forecast. Finally, end-use analysis, without an
projected number of appliances in each home econometric component that is explained above,
IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 5, No 3, September 2011
ISSN (Online): 1694-0814
www.IJCSI.org 511
does not take price changes (elasticity of defined in terms of a multitude of factors (policy
demand) in electricity or other competing fuels factors, price factors, end-use factors), it is
into consideration. flexible and useful for analyzing load growth
under different scenarios.
Ideally this approach is very accurate. However,
it is sensitive to the amount and quality of end- A disadvantage of econometric forecasting is
use data. For example, in this method the that in order for an econometric forecast to be
distribution of equipment age is important for accurate, the changes in electricity demand
particular types of appliances. End-use forecast caused by changes in the factors influencing that
requires less historical data but more information demand must remain the same in the forecast
about customers and their equipment [1]. period as in the past. This assumption (which is
called constant elasticities) may be hard to
5.3 Econometric justify, especially where very large electricity
price changes (as opposed to small, gradual
Econometrics uses economics, mathematics, and changes) make consumers more sensitive to
statistics to forecast electricity demand. electricity prices [3].
Econometrics is a combination of trend analysis
and end-use analysis, but it does not make the Also, the econometric load forecast can only be
trend-analysts assumption that future electricity as accurate as the forecasts of factors which
demand can be projected based on past demand. influence demand. Because the future is not
Moreover, unlike many end-use models, known, projections of very important demand-
econometrics can allow for variations in the influencing factors such as electricity, natural
relationship between electricity input and end- gas, or oil prices over a 10- or 20-year period
use. are, at best, educated guesses. Finally) many of
the demand-influencing factors which may be
Econometrics uses complex mathematical treated and projected individually in the
equations to show past relationships between mathematical equations could actually depend on
electricity demand and the factors which each other, and it is difficult to determine the
influence that demand. For instance, an equation nature of these interrelationships. For example,
can show how electricity demand in the past higher industrial electricity rates may decrease
reacted to population growth, price changes, etc. industrial employment, and projecting both of
For each influencing factor, the equation can them to increase at the same time may be
show whether the factor caused an increase or incorrect. A model which treats projected
decrease in electricity demand, as well as the size industrial electricity rates and industrial
(in percent) of the increase or decrease. For price employment separately would not show this fact.
changes, the equation can also show how long it
took consumers to respond to the changes. The Econometric models work best when forecasting
equation is then tested and fine tuned to make at national, regional, or state levels. For smaller
sure that it is as reliable a representation as geographical areas, meeting the model can be a
possible of the past relationships. Once this is problem. This is oddly shaped service areas for
done, projected values of demand-influencing which there demographic data.
factors (population, income, prices) are put into
the equation to make the forecast. A similar 6.0 COMPARISON OF
procedure is followed for all of the equations in ELECTRICAL LOAD
the model. FORECASTING
TECHNIQUES
The advantages of econometrics are that it
provides detailed information on future levels of In the previous discussion we focus on electrical
electricity demand, why future electricity load forecasting techniques, most forecasting
demand increases or decreases, and how methods use statistical techniques or artificial
electricity demand is affected by various factors. intelligence algorithms such as regression, neural
In addition, it provides separate load forecasts networks, fuzzy logic, and expert systems. Two
for residential, commercial, and industrial of the methods named trend analysis, end-use
sectors. Because the econometric model is and econometric approach are broadly used for
IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 5, No 3, September 2011
ISSN (Online): 1694-0814
www.IJCSI.org 512
medium- and long-term forecasting. A variety of reason, explain, and have its knowledge base
methods, which include the similar day expanded as new information becomes available
approach, various regression models, time series, to it. The load forecast model is built using the
neural networks, statistical learning algorithms, knowledge about the load forecast domain from
fuzzy logic, and expert systems, have been an expert in the field. This knowledge is
developed for short-term forecasting. represented as facts and rules by using the first
predicate logic to represent the facts and IF-
The method for short-term forecasting are THEN production rules. This representation is
similar day approach, various regression models, built in what is called the knowledge base
time series, neural networks, statistical learning component of the expert system. The search for
algorithms, fuzzy logic, and expert systems. solution or reasoning about the conclusion drawn
Similar day approach is based on searching by the expert system is performed by the
historical data of days of one, two or three years "Inference Engine" component of the expert
having the similar characteristics to the day of system. For any expert system it has to have the
forecast. Regression is the one of most widely capability to trace its reasoning if asked by the
used statistical techniques. For electric load user. This facility is built through an explanatory
forecasting, regression methods are usually used interface component. Fuzzy logic based on the
to model the relationship of load consumption usual Boolean logic which is used for digital
and other factors such as weather, day type, and circuit design. In case of fuzzy logic, the input is
customer class. There are several regression related to the comparison based on qualities. The
models for the next day peak forecasting. Their advantage of fuzzy logic is that there is no need
models contain deterministic influences such as of mathematical models for mapping between
holidays, random variables influences such as inputs and outputs and also there is no need of
average loads, and exogenous influences such as precise or even noise free inputs. Based on the
weather. Time series is a very popular approach general rules, properly designed fuzzy logic
for the electrical load forecasting. Two important systems are very strong for the electrical load
models of time series are ARMA and ARIMA. forecasting.
ARMA and ARIMA use the time and load as the
only input parameters. Since load generally The methods for long- and medium-term
depends on the weather and time of the day, forecasting are trend analysis, end-use and
ARIMAX is the most natural tool for load econometric approach. The advantage of trend
forecasting among the classical time series analysis is that it is quick, simple and
models [2]. inexpensive to perform and does not require
much previous data. The basic idea of the end-
The other methods are based on Artificial use analysis is that the demand for electricity
intelligence, they are called Intelligent Systems. depends what it use for (the end-use). The
In Artificial Neural Network, forecasting is advantages of end-use analysis is that it
based on the pattern observed from the past identifies exactly where electricity goes, how
event and estimates the values for the future. much is used for each purpose, and the potential
ANN is well suited to forecasting for two for additional conservation for each end-use. The
reasons. First, it has been demonstrated that disadvantage of end-use analysis is that most
ANN are able to approximate numerically any end-use models assume a constant relationship
continuous function to be desired accuracy. In between electricity and end-use (electricity per
this case the ANN is seen as multivariate, appliance, or electricity used per dollar of
nonlinear and nonparametric methods. Secondly, industrial output). This might hold true over a
ANNs are date-driven methods, in the sense that few years, but over a 10-or 20-year period,
it is not necessary for the researcher to use energy savings technology or energy prices will
tentative modals and then estimate their undoubtedly change, and the relationships will
parameters. ANNs are able to automatically map not remain constant. The advantages of
the relationship between input and output, they econometrics are that it provides detailed
learn this relationship and store this learning into information on future levels of electricity
their parameters. An Expert System is a demand, why future electricity demand increases
computer program, which has the ability to act as or decreases, and how electricity demand is
an expert. This means this computer program can affected by various factors. A disadvantage of
IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 5, No 3, September 2011
ISSN (Online): 1694-0814
www.IJCSI.org 513
REFERENCES