Documentos de Académico
Documentos de Profesional
Documentos de Cultura
CHAPTER 1
INTRODUCTION
Forecasting is the process of making statements about the events whose actual
outcomes has not been yet observed. It is a systematic effort to anticipate future events
and conditions. As power system planning and construction requires a gestation period
of four to eight years or even longer for the present day super power stations energy
and load demand forecasting plays an important role. Related to Power system
forecasting means estimating active load at various load buses ahead of actual load
occurrences. The importance of demand forecasting needs to be emphasized at all
levels as the consequences of under or over forecasting the demand are serious and will
affect all stakeholders in the electricity supply industry. If under estimated, the result is
serious since plant installation cannot easily be advanced, it will affect the economy,
business, loss of time and image. If over estimated, the financial penalty for excess
capacity (i.e., over-estimated and wasting of resources). Nature of forecasts, lead times
and application are summarized in the table.
TABLE 1. TYPES OF FORECAST AND ITS APPLICATION
Nature of forecast Lead time Application
Very short term A few seconds to several Generation, distribution
minutes schedules. contingency
analysis for system security
Short term Half an hour to a few hours Allocation of spinning
reserve, operational planning
and unit commitment,
maintanence and scheduling
Medium term A few days to few weeks Planning for seasonal peak
winter, summer
Long term A few months to a few years Planning generation growth
Two approaches used for load forecasting is total load approach and component
approach. Advantage of total load approach is, it is much smoother and indicative of
overall growth trends and it is easy to apply. On the other hand component approach
shows the abnormal conditions in growth trends of a certain component which prevents
the misleading forecast conclusions. All the forecasting techniques are based on the
assumption that the load supplied will meet the system demand at all points of time. A
statistical analysis of previous load data is used to set up a demand pattern. Once this
has been done, this load model is used for predicting the estimated demand for selected
time. The major task associated with forecasting is to select the best load model this can
be done by decomposing the given load demand at any given point of time into number
of distinctive components. The load is depending on industrial, commercial and
agricultural activities and the weather conditions like temperature, cloudiness, wind
velocity, visibility and precipitation.
Since there are few comparative studies available to help the forecaster to choose
wisely among the various forecasting schemes, some of which are already in use and
some of which are not yet developed to the point of practical implementation. This
paper reviews six possible approaches to peak-demand forecasting but places greatest
emphasis on probabilistic monthly or weekly peak demand forecasting. Due to the vast
extent of relevant literature, only a few contributions to weather sensitive peak
forecasting, will be pointed out here. A methodology for forecasting annual or weekly
and monthly peak loads, based on the decomposition of peak loads into nonweather
sensitive and weather sensitive components, was developed in [1], [2] and [3]. In [4], a
weather-sensitive model for summer afternoon peak loads, applied to both long and
short term forecasting, using a normal distribution was developed. In [5], a different
regression model, used in forecasting is described. This methodology was modified in
order to apply for a practical system in [6].
CHAPTER 2
Energy forecasting can be done with more reliability than demand forecasts. Annual
energy can be converted into demand by multiplying it with load factor. Therefore it is
possible to obtain a demand forecast by combining the forecasting of energy and load
factor which is superior in accuracy to a peak demand forecasted by direct methods. It
is easy to obtain energy forecasting because of the availability of data but forecasting of
load factor is just as difficult as the annual peak demand. The accuracy of load factor
forecast can be maintained but the work involved in forecasting is comparable with
peak demand forecasting.
Advantage of energy forecasting is the availability of data from which the energy
consumption of different classes of customers can be known.
Limitation of this method is that it does not provides estimates of the variance of
annual peak demand forecasts. This method is used for finding the seasonal peak
demands.
weekly peak demand data is more suitable for obtaining weather load model it is used
over monthly peak demand data also a week is a precisely defined time period whereas
a month possesses an irregular number of potential peak demand days. The procedure is
similar with that of the method specified by the 4 th method but in method describes
above it is assumed that the weather sensitive component is normally distributed but in
actual case it is not strictly normally distributed and it gives no indication of the
difficulties caused by the non Gaussian distribution of the weather sensitive component
of weekly peak demand. If it is assumed that the non- weather sensitive component of
weekly peak demand is normally distributed, and that its variance is several times the
variance of weather sensitive component, it is a reasonable approximation to assume
that the total load is normally distributed. The total peak demand forecast at some
future time has a mean value
D (t) = B (t) + S (t)
Where B (t) is the mean value of the non-weather sensitive component of weekly peak
demand and S (t) is the mean value of the weather sensitive component of the weekly
peak demand. Its variance is
σ2 D(t) = σ2 B(t) + σ2 S(t)
Although these results can be applied to any distribution, the usefulness of variance by
setting up a confidence interval for forecasts does depends on the total forecast being
nearly Gaussian.
Having obtained a weekly peak demand methods should be formulated to extract
information concerning the annual or seasonal peak demands. It should be noted that
the peak demand forecasted for a particular week is not a number rather a probability
distribution. Although we can represent the distribution by means of its mean and
variance but these two variables is not sufficient to describe the distribution. It is
temporarily assumed that the distribution of weekly peak demand is known and is not
necessarily Gaussian. Let Pk(x) be the cumulative probability distribution of peak
demand of kth week and pk(x) be the probability density function of kth week. The
problem is to determine the annual or seasonal peak demand from 52 different
probability distribution. The probability density function for annual or seasonal peak
demand is given by
Where N=52 for annual peak and N=26 for seasonal peak. The result is quite general
and time consuming to evaluate using a digital computer. This can be simplified by
considering only those weeks where annual or seasonal peak occurs, in this way we can
reduce the value of N there by reducing the computational effort.
Procedure can be summarized as follows
1. Forecast both mean and peak demand for each week
2. Assume the weekly peak demand is normally distributed and calculate the
probability density function of annual peak demand that includes only those
weeks in which annual peak demand occurs.
3. Calculate mean and variance of the annual peak demand using
Where Xt is the estimate of monthly peak demand produced by the model, ut, vt
and wt are variables with zero mean, Gaussian random process, ρ is a parameter
varying between 0 and 1, and U is the backward shift operator. Four parameters
must be estimated before the model is complete, the variance of random inputs σu2,
σv2, σw2 and a parameter of the model ρ. However no meaningful statistics can be
developed from historical data due to non stationarity so a reversible transformation
which converts the non stationary process Xt to stationary process Yt.
CHAPTER 3
Regression model
Regression model involves three variables
1. Unknown parameter denoted as β which may be a scalar or a vector
2. Dependent variable, Y
3. Independent variable, X
Assume now that the vector of unknown parameters β is of length k. In order to perform a
regression analysis the user must provide information about the dependent variable Y:
• If N data points of the form (Y, X) are observed, where N < k, most classical
approaches to regression analysis cannot be performed: since the system of
equations defining the regression model is underdetermined, there is not enough
data to recover β.
• If exactly N = k data points are observed, and the function f is linear, the
equations Y = f(X, β) can be solved exactly rather than approximately. This
reduces to solving a set of N equations with N unknowns (the elements of β),
When the number of measurements, N, is larger than the number of unknown parameters,
k, and the measurement errors εi are normally distributed then excess of information
contained in (N - k) measurements is used to make statistical predictions about the
unknown parameters. This excess of information is referred to as the degrees of freedom
of the regression.
Classical Assumptions
1. LINEAR REGRESSION
there may be relationship between three variables X, Y, and Z that can be described
by the equation
Z= β0 + β1 X+ β2 Y, which is called a linear equation in the variables X, Y, and Z. In a
three dimensional rectangular coordinate system this equation represents a plane.
2. QUADRATIC REGRESSION
Quadratic Regression is in the form of a parabola that means the function will
first increase then decrease or first decrease then increase. It can be described by the
equation
3. CUBIC REGRESSION
Cubic regression is described by the equation
Y= β3 X3 + β2 X2 + β1 X+ β0
Cubic regression can increase then decrease and then increase or it will decrease then
increase and then decrease.
4. QUARTIC REGRESSION
It is described by the equation
Y= β4 X4+ β3 X3 + β2 X2+β1 X+
Quartic regression can increase then decrease then increase then decrease or it will
decrease then increase then decrease then increase.
5. EXPONENTIAL REGRESSION.
It is described by the equation
Y= β0 + β1x . It will strictly increase or decrease.
The demand data may be prepared on daily, weekly or hourly basis it may consists of
several useful variables, it is desirable to include only the most promising variables on
the data so that subsequent experimentation is possible without reprocessing the hourly
data. The results were produced with daily and weekly peak demand and coincident dry
bulb temperature.
of linear regression to fit kw and Ks values of Tw and Ts have been specified. The
specific procedure is as follows.
D0 is the average value of all the demand values lying in the temperature range of
Tw<T<Ts. Since the winter weather load characteristic must pass through the point
(D0, Tw) a least squares, straight line fit gives the value of Kw. Similarly, the summer
characteristics passes through the point (D0, Ts) and Ks is readily determined by a least
squares fit. Since the points (D0, Kw) and (D0, Ks) are already specified some data
points are excluded while estimating the values of Kw and Ks. All heating days with
peak demand less than D0 are excluded while fitting Ks and all cooling days with peak
demand less than D0 are excluded while fitting Ks.
The values of the base load obtained in the step 2 is fitted to a suitable trend curve
and are extrapolated and thus obtaining the mean value and variance of the base load at
every month of the forecasting period. A linear trend curve is utilized for this purpose.
Variation of base load from month to month for every year of historical data annual
behaviour pattern was observed. In the forecasted values, monthly coefficients are
calculated and then imposed on the values of the base load resulting from the
extrapolation.
In this step, the morning and afternoon weather-load models created in step 1 are
extrapolated resulting in the respective models for every year of the forecast period.
More specifically, the forecasted weather-load models are created by extrapolating the
value of the first derivative of the historical models at various temperatures. Particular
care must be given to the impact of the level of saturation of heating or cooling devices.
that the temperature during peak hours and the weather sensitive load component are
normally distributed.
CHAPTER 4
RESULTS
The above method is applied to Hellenic system. Historical data for the peak load
and the temperature variation is used. The mean monthly peak loads are
decomposed into base load and weather sensitive load components. The known
weather load models and base loads are extrapolated. The extrapolation result is as
shown in the figure4.1
CHAPTER 5
CONCLUSION
The implementation of forecasting technique “Separate treatment of weather
sensitive component” is described in the paper which is quite straight forward and
also a widely used technique. The theoretical implementation of this load
forecasting approach for real system is not so straight forward so the method with
little modification is described here. The accuracy of this method is acceptable to
some extend. The method uses statistical extrapolation technique with lots of
assumptions which suffers from following drawbacks like annual peak load data of
particular element for example substation, can deviate much during the loading
history. These data shifts are mostly caused by switching as loads are routinely
moved from one substation to another during the course of utility operation. Hence
the extrapolation technique suffers badly due to these data shifts. This problem can
be significant, especially in electricity distribution systems with an incomplete
Remote control Systems. In order to solve these problems improve techniques has
to be used. Forecasting using neural networks which models all these uncertainties
is recommended.
APPENDIX 1
where dY/dC is the value of the derivative of the weather-load model at temperature
and
REFERENCES
[1] K.N Stanton and Pradeep C Guptha, “Forecasting Annual or seasonal peak demand
in Electric utility systems” IEEE Transactions of Power Apparatus and systems pp.951
– 959 May/June 1970
[2] K.N Stanton, “Medium Range, Weekly and Seasonal Peak Demand Forecasting by
Probability methods” IEEE Transactions pp 1183-1189June 1970”
[3] R.L Sullivan, “Power System Planning”, Tata Mcgraw Hill, Newyork. 1977, pp 19-
59
[4] C.E Asburgy, “Weather load model for electric demand and energy forecasting”
IEEE Transactions on Power Apparatus and Systems, vol.-94, no.4 July/August 1975
[5] Murray R. Spielgal and Larry J.Stephens, “Shaum’s Outline of Statistics” , Mcgraw
Hill ,Newyork , 1998 pp.282-286,311-318,434-439
[6] E.G Contaxi, “Application of weather sensitive Peak load forecasting Model to the
Hellenic system” IEEE Melecon 2004 pp.819-822
[7] Slobodan M. Maksimovich and Vladimir M. Shiljkut, “The Peak load forecasting
afterwards its intensive reduction” IEEE Transactions on power delivery, vol, 24, N0.3,
July 2009 pp 1552-1559