A

Seminar Report on
Short Term Price Forecasting
Submitted in partial fulfillment of the requirement for the
degree of

BACHELOR OF TECHNOLOGY
In
ELECTRICAL ENGINEERING

BY
PRAVESH SRIVASTAVA
(Roll No.-14335)
Under the guidance of
Mr. ANKIT KUMAR SRIVASTAVA

Department of Electrical Engineering

Kamla Nehru Institute of Technology, Sultanpur

Dr. A. P. J. ABDUL KALAM TECHNICAL UNIVERSITY, LUCKNOW

(1 )
 Electrical Engineering Department
Kamla Nehru Institute of Technology,Sultanpur-228118(U.P.)
India

CERTIFICATE
This is certified that seminar report on SHORT TERM PRICE FORECASTING submitted
by Mr. PRAVESH SRIVASTAVA, B. Tech., Department of Electrical Engineering, Kamla
Nehru Institute of Technology, Sultanpur, for the partial fulfillment of the award of degree of
Bachelor of Technology affiliated to A.P.J Abdul Kalam Technical University, Lucknow,
under my supervision during the academic session 2016-17. The seminar embodied results of
original work, and studies are carried out by the student himself and the contents of the seminar
do not form the basis for the award of any other degree to the candidate or to anybody else from
this or any other University/Institution.

SUPERVISOR

(Mr. Ankit Kumar Srivastava)

APPROVAL FOR SUBMISSION

Seminar Coordinator (Dr. A.S. Pandey)

(Prof. Varun Kumar) Professor & Head

(Professor) Department of Electrical Engineering

Kamla Nehru Institute of Technology,

Sultanpur-228118(U.P.) India

(2 )
 ACKNOWLEDGEMENT

I would like to express my deepest gratitude to Dr. A. S. Pandey the head of the
department, Prof. Varun Kumar the seminar in-charge and Mr. Ankit Kumar Srivastava
the seminar guide who sincerely extended their help and provided resourceful and helpful
inputs without which the work would never have been accomplished.
I extend my cordial gratitude and esteem to my teachers, whose effective guidance, valuable
time and constant inspiration made it feasible and easy to carry out the work in a smooth
manner.
I express my gratitude to all my friends and classmates for their support and help in this
seminar. Last, but not the least I wish to express my gratitude to almighty God for his
abundant blessings, without which this seminar would not have been successful.

Pravesh Srivastava
B. Tech 6th Semester
Electrical Engineering
K.N.I.T Sultanpur
Uttar Pradesh

(3 )
 CONTENT
1. ABSTRACT (5)

2. INTRODUCTION (6)

3. BACKGROUND (8)

4. PRICE-FORECASTING METHODOLOGIES (9)

4.1. GAME THEORY MODEL (9)

4.2. SIMULATION MODELS (10)

4.3 TIME SERIES MODELS (11)

4.3.1. PARSIMONIOUS STOCHASTIC MODELS (11)

4.3.1.1Auto Regressive Model (12)

4.3.1.2 Moving Average Model (13)

4.3.1.3ARMA Model (13)

4.3.1.4ARIMA Model (14)

4.3.1.5GARCH Model (16)

4.3.2. REGRESSION OR CAUSAL MODELS (17)
4.3.3ARTIFICIAL INTELLIGENCE (AI) MODELS (17)
4.3.3.1 ANN based models (17)
4.3.3.2Data-mining models (23)
4.3.3.3.1Fuzzy Model (23)
5. FACTORS INFLUENCING ELECTRICITY PRICES (25)

(4 )
 1. ABSTRACT

Since early 1980's the electrical industry is going under a continuous change. This change is
leading to a complete different atmosphere where the ultimate benefit is provided to the end use
customers, with a reliable and cheap electricity supply. This market is a customer driven market
and price forecasting is an important tool to the market players. Various models and techniques
have been developed by the researchers to determine the correct price in order to obtain the
maximum profit. Discussion on various price forecasting techniques and their application in
various electricity markets around the world has been discussed .The inculcation of heuristic and
non-heuristic methods brings into development of price forecasting techniques with involvement
of advantages and drawbacks brings about the interesting prospect of the topic in this seminar
study.
There has been various techniques and tools developed for price forecasting and it brings
into account a broader classification of the study. The methodologies used is divided into three
models:

1. Game theory Model.

2. Time Series Model.
3. Simulation Model.
The 2nd model is yet further broadly divided and most of the techniques used comes into this
section. The various techniques being used in this are ANN, ARIMA, Hybrid model, and data
mining model.
The time series model being used deploys easy tool and techniques for forecasting and bringing
out the linear and non-linear relationships among the input and output variables.

(5 )
 2. INTRODUCTION

Electricity price forecasting has been one of the important and interesting aspect, keeping into its
need for the public welfare including both consumer and producer. The cost shifting for
maximizing profit and at the same time providing reasonable and affordable access to bill for
unit has been the agenda of price forecasting. An accurate day-ahead price forecast in the market
helps the power suppliers adjust their bidding strategies to achieve maximum benefit. On the
other hand, suppliers can derive a plan to maximize their utilities using electricity purchased
from the pool, or they can use self-production capacity to protect themselves against high prices.
However, electricity has distinct characteristics from other commodities. Electrical energy cannot
be effectively stored and the need for stability of the power system demands a continual
balancing of generation and load. On short timescales, most users of electricity are unaware of,
or indifferent to, its price. Transmission bottlenecks usually limit electricity transportation from
one region to another. These factors create the extreme price volatility, and even price spikes, in
the electricity market.
The power industry across the globe is experiencing a major change in its business as
well as in an operational model where, the vertically integrated utilities are being unbundled and
opened up for competition with private players putting an end to the era of monopoly. Today
energy price forecasting is an important area of research. Market participants of this market such
as generators, power suppliers, investors and trades need to maximize their profit.
Under restructuring of electric power industry, different participants namely generation
companies and consumers of electricity need to meet in a marketplace to decide on the electricity
price. In the current deregulated scenario, the forecasting of electricity demand and price has
emerged as one of the major research fields in electrical engineering. A lot of researchers and
academicians are engaged in the activity of developing tools and algorithms for load and price
forecasting. Whereas, load forecasting has reached advanced stage of development and load
forecasting algorithms with mean absolute percentage error (MAPE) below 3% are
available, price-forecasting techniques, which are being applied, are still in their early
stages of maturity. In actual electricity markets, price curve exhibits considerably richer
structure than load curve and has the following characteristics: high frequency, non-constant
mean and variance, multiple seasonality, calendar effect, high level of volatility and high

(6 )
percentage of unusual price movements. All these characteristics can be attributed to the
following reasons, which distinguish electricity from other commodities:
(i) Non-storable nature of electrical energy.
(ii) The requirement of maintaining constant balance between demand and
supply.
(iii) Inelastic nature of demand over short time period.
(iv) Oligopolistic generation side.

In addition to these, market equilibrium is also influenced by both load and generation side
uncertainties. Therefore price-forecasting tools are essential for all market participants for their
survival under new deregulated environment. Even accurate load forecasts cannot guarantee
profits and the market risk due to trading is considerable because of extreme volatility of
electricity prices.
The electricity price can rise to tens or even hundreds of times its normal value, and so is one of
most volatile of all commodities The importance and complexity of electricity price forecasting
has motivated a great deal of research in this area, especially in recent years. Stationary time
series models, such as autoregressive, dynamic regression and transfer function and
autoregressive integrated moving average (ARIMA) models have been proposed for this
purpose. Also non-stationary time series models such as the generalized autoregressive
conditional heteroskedastic (GARCH) model have been used too. However, most time series
models are linear predictors, while electricity prices are generally nonlinear functions of their
inputs. Therefore, the behavior of price signals may not be completely captured by these time
series techniques. To solve this problem, some other research has proposed the use of neural
networks and fuzzy neural networks for price forecasting. Neural networks and fuzzy neural
networks can model non-linear input/output mapping functions. However, the price of electricity
is a time-varying signal and its functional relationships vary rapidly with time. So, information
derived from the neural networks or fuzzy neural networks rapidly loses its value. While it
appears to be the case that neural networks and fuzzy neural networks learn well from training
data, they may exhibit large prediction errors in the test phase.

(7 )
 3. BACKGROUND
Since the early 1990s, the process of deregulation and the introduction of competitive electricity
markets have been reshaping the landscape of the traditionally monopolistic and government-
controlled power sectors. Throughout Europe, North America and Australia, electricity is now
traded under market rules using spot and derivative contracts. However, electricity is a very
special commodity: it is economically non-storable and power system stability requires a
constant balance between production and consumption. At the same time, electricity demand
depends on weather (temperature, wind speed, precipitation, etc.) and the intensity of business
and everyday activities (on-peak vs. off-peak hours, weekdays vs. weekends, holidays, etc.).
These unique characteristics lead to price dynamics not observed in any other market, exhibiting
daily, weekly and often annual seasonality and abrupt, short-lived and generally
unanticipated price spikes.

Extreme price volatility, which can be up to two orders of magnitude higher than that of any
other commodity or financial asset, has forced market participants to hedge not only volume but
also price risk. Price forecasts from a few hours to a few months ahead have become of particular
interest to power portfolio managers. A power market company able to forecast the volatile
wholesale prices with a reasonable level of accuracy can adjust its bidding strategy and its own
production or consumption schedule in order to reduce the risk or maximize the profits in day-
ahead trading. A ballpark estimate of savings from a 1% reduction in the mean absolute
percentage error (MAPE) of short-term price forecasts is $300,000 per year for a utility with
1GW peak load.

(8 )
 4. PRICE-FORECASTING METHODOLOGIES
Numerous methods have been developed for electricity price forecasting. Time horizon varies
from hour ahead to a week ahead forecasting. The price-forecasting models have been classified
in three sets and these three sets have been further divided into subsets as shown in Fig. 1.

Electricity Price Forecasting Techniques

Game Theory Models Time Series Model Simulation Model

Parsimonious Artificial Intelligence Regression or Causal

Stochastic Models Based models Models

Data-mining Models
Neural Network
Based Models

Fig-1

4.1. GAME THEORY MODEL

The first group of models is based on game theory. It is of great interest to model the strategies
(or gaming) of the market participants and identify solution of those games. Since participants in
oligopolistic electricity markets shift their bidding curves from their actual marginal costs in
order to maximize their profits, these models involve the mathematical solution of these games
and price evolution can be considered as the outcome of a power transaction game. In this group
of models, equilibrium models take the analysis of strategic market equilibrium as a key point.
There are several equilibrium models available like Nash equilibrium, Cournot model, Bertrand
model, and supply function equilibrium.

(9 )
In China semi-deregulated electricity market, a game-theoretic model has been set between users
and power suppliers in order to ensure the whole network voltage load curve stay smooth, as
well as to make the suppliers profitable on the basis of the satisfaction of users in power
utilization in daily lives. The foundation of the model is the definition of the price elasticity of
demand and its characters in micro-economy. According to this model, a Bayesian Nash
equilibrium has been put forward based on the demand price elasticity, which can not only
smooth the whole network voltage load curve effectively through the reciprocal process between
users and power suppliers, but also can optimize the profit of the power suppliers. Study of game
theory models in itself is a major area of research and has been kept outside the scope of this
study.

4.2. SIMULATION MODELS

These models form the second class of price-forecasting techniques, where an exact model of the
system is built, and the solution is found using algorithms that consider the physical phenomenon
that governs the process. Then, based on the model and the procedure, the simulation method
establishes mathematical models and solves them for price forecasting.
Price forecasting by simulation methods mimics the actual dispatch with system
operating requirements and constraints. It intends to solve a security constrained optimal power
flow (SCOPF) with the entire system range. Two kinds of simulation models have been
analyzed. One is market assessment and portfolio strategies (MAPS) algorithm developed by GE
Power Systems Energy Consulting and the other is UPLAN software developed by LCG
Consulting. MAPS is used to capture hour-by-hour market dynamics while simulating the
transmission constraints on the power system. Inputs to MAPS are detailed load, transmission
and generation units data. Whereas the outputs are complete unit dispatch information, LMP
prices at generator buses, load buses and transmission flow information. UPLAN, a structural
multi-commodity, multi-area optimal power flow (MMOPF) type model, performs Monte Carlo
simulation to take into account all major price drivers. UPLAN is used to forecast electricity
prices and to simulate the participants behavior in the energy and other electricity markets like
ancillary service market, emission allowance market. The inputs to MMOPF are competitive
bidding behavior, generation units data, the transmission network data, hydrological conditions,

( 10 )
fuel prices and demand forecasts. These are almost comparable to the input variables of MAPS.
The outputs are forecast of prices and their probability distribution across different energy
markets. The dynamic effect of drivers on market behavior has also been captured.
Both UPLAN and MAPS may be used for long as well as short range planning. Simulation
methods are intended to provide detailed insights into system prices. However, these methods
suffer from two drawbacks. First, they require detailed system operation data and second,
simulation methods are complicated to implement and their computational cost is very high.

4.3 TIME SERIES MODELS

Time series analysis is a method of forecasting which focuses on the past behavior of the
dependent variable. Sometimes exogenous variables can also be included within a time series
framework.
Short-term price forecasting presents a crucial activity for the profitability of generation
companies, end-user companies and trading companies. It enables them to optimize their trading
strategies. The planning horizon for generation and consumer companies typically spans weeks
and/or months. Thus there is a need for models that can forecast the average weekly spot prices.
Time series models represent an important category of models. These models typically utilize
some external variables related to supply and/or demand to explain pricing behavior. Other
approaches include multiple support vector machine, extreme learning machine, auto-regressive
fractionally integrated moving average and Grey models. Time series models are among the
models with strongest forecasting ability. Single hour model specifications achieve better
forecasting accuracy than multi-hour model specifications.

4.3.1. PARSIMONIOUS STOCHASTIC MODELS

Many stochastic models are inspired by the financial literature and a desire to adapt some of the
well-known and widely applied in practice approaches. Univariate discrete type models like
autoregressive (AR), moving average (MA), autoregressive moving average (ARMA),
autoregressive integrated moving average (ARIMA), and generalized autoregressive conditional

( 11 )
heteroskedastic (GARCH) are considered. These are discrete time counterparts corresponding to
the continuous-time stochastic models. Purely finance-inspired stochastic models involving
certain characteristics of electricity prices, like price spikes and mean reversion, have been kept
outside the scope of this review. A discussion on these models can be seen in. Stochastic time
series can be divided into stationary process and non-stationary process. The basic assumption of
stationarity on the error terms includes zero mean and constant variance. In AR, MA and ARMA
models conditions of stationarity are satisfied; therefore they are applicable only to stationary
series. ARIMA model tries to capture the incremental evolution in the price instead of the price
value. By the use of a difference operator, transformation of a non-stationary process into a
stationary process is performed.
The class of models where the constant variance assumption does not need to hold is
named heteroskedastic. Thus GARCH mode considers the conditional variance as time
dependent. In all these models price is expressed in terms of its history and a white noise
process. If other variables are affecting the value of price, the effect of these variables can be
accounted for using multivariate models like TF (transfer function) and ARMA with exogenous
variables (ARMAX) models. As electricity price is a non-stationary process, which exhibits
daily, weekly, yearly and other periodicities. Therefore, a different class of models that have this
property, designated as seasonal process model, is used.

4.3.1.1Auto Regressive Model

In statistics and signal processing, an autoregressive (AR) model is a representation of a type

of random process; as such, it is used to describe certain time-varying processes
in nature, economics, etc. The autoregressive model specifies that the output variable
depends linearly on its own previous values and on a stochastic term (an imperfectly predictable
term); thus the model is in the form of a stochastic difference equation.
The notation AR (p) refers to the autoregressive model of order p. The AR (p) model is written
p
X t=c + i X ti+ t
i=1

( 12 )
where 1,,,, p are parameters, c is a constant, and the random variable t is white noise.
Some constraints are necessary on the values of the parameters so that the model
remains stationary. For example, processes in the AR (1) model with |1| 1 are not stationary.

4.3.1.2 Moving Average Model

In time series analysis, the moving-average (MA) model is a common approach for
modeling univariate time series. The moving-average model specifies that the output variable
depends linearly on the current and various past values of a stochastic (imperfectly predictable)
term.

The notation MA (q) refers to the moving average model of order q

q
X t=+ t + i ti
i=1

Where the 1 ,... , q are the parameters of the model, is the expectation of Xt (often

assumed to equal 0), and the t { varepsilon{t }}, t i ,... are again, white noise error terms.

4.3.1.3ARMA Model

In time series analysis, the moving-average (MA) model is a common approach for
modeling univariate time series. The moving-average model specifies that the output variable
depends linearly on the current and various past values of a stochastic (imperfectly predictable)
term.

Together with the autoregressive (AR) model, the moving-average model is a special case
and key component of the more general ARMA and ARIMA models of time series, which have a
more complicated stochastic structure

The notation ARMA (p, q) refers to the model with p autoregressive terms and q moving-
average terms. This model contains the AR (p) and MA (q) models,

( 13 )
 pq
X t=c + i X ti+ t+ i ti

i=1 i=1

4.3.1.4ARIMA Model

Autoregressive moving average (ARMA) model introduced by Box and Jenkins is a stochastic
process that is built-in to time series data to get more understanding of the data and to do future
predictions in the series. Non seasonal ARIMA is denoted by (x,y,z) in which x, y, and z are
positive integers, where x is defined as order of the Autoregressive model, y degree of
differencing, and z is the order of the Moving-average model. In Seasonal ARIMA models are
given by ARIMA (x,y,z) (X,Y,Z)n where n is number of periods in each season. X,Y,Z refers to
the autoregressive, differencing, and moving average terms for the seasonal part of the ARIMA
model. These models provide reliable and accurate forecasting.
ARIMA models have already been used for forecasting of oil, gas, and natural as well as
load forecasting in power system with good results. Some examples of energy market where this
model is being used. Simpler version of ARIMA Auto Regressive (AR) models have already
been being used in Norwegian system for forecasting weekly prices. ARIMA model is also used
in California and mainland Spain markets and satisfactory results are obtained. The use of
ARIMA Models along with wavelet transform was used to predict price in mainland Spain for
year 2000 was done. For Californian power market also the ARIMA models was used for
forecasting daily average prices, based on historical data was done.

Given a time series of data Xt where t is an integer index and the Xt are real numbers, an
ARMA(p,q) model is given by

( 14 )
or equivalently by

where L is the lag operator, the { alpha i }} i are the parameters of the autoregressive part of
{

the model, the { theta i }} i are the parameters of the moving average part and
{

the { varepsilon t }} t are error terms. The error terms { varepsilon t }} t are generally
{ {

assumed to be independent, identically distributed variables sampled from a normal

distribution with zero mean.

An ARIMA (p, d,q) process expresses this polynomial factorization property with p=p'd, and is
given by:

and thus can be thought as a particular case of an ARMA(p+d,q) process having the
autoregressive polynomial with d unit roots. (For this reason, no ARIMA model with d > 0
is wide sense stationary.)The above can be generalized as follows.

( 15 )
4.3.1.5 GARCH Model

Electricity price is so unpredictable, that price spikes are surely to happen this makes forecasting
difficult. In order to solve this problem, we put up another model that is generalized
autoregressive conditional heteroskedasticity GARCH, can resolve this volatility problem. It
considers the conditional variance as time dependent hear price is expressed in terms of its
history and a white noise process. Some case studies have been made which uses GARCH
method for price forecasting with the aim to model the dynamic character of price. A
comprehensive explanation of the GARCH models and its application in mainland Spain and
Californian markets are presented in paper and also examined on Spanish and PJM electricity
markets along with the comparison with other forecasting methods.

If an autoregressive moving average model (ARMA model) is assumed for the error
variance, the model is a generalized autoregressive conditional heteroskedasticity (GARCH)
model.

In that case, the GARCH (p, q) model (where p is the order of the GARCH terms 2 and q is
the order of the ARCH terms 2 ), is given by

( 16 )
Generally, when testing for heteroskedasticity in econometric models, the best test is the White
test. However, when dealing with time series data, this means to test for ARCH and GARCH
errors.

4.3.2. REGRESSION OR CAUSAL MODELS

Regression type forecasting model is based on the theorized relationship between a dependent
variable (electricity price) and a number of independent variables that are known or can be
estimated. The price is modeled as a function of some exogenous variables. The explanatory
variables of this model are identified on the basis of correlation analysis on each of these
independent variables with the price (dependent) variable.

4.3.3ARTIFICIAL INTELLIGENCE (AI) MODELS

These may be considered as nonparametric models that map the inputoutput relationship
without exploring the underlying process. It is considered that AI models have the ability to learn
complex and nonlinear relationships that are difficult to model with conventional models. These
models can be further divided into two categories:
(i) Artificial neural network (ANN) based models
(ii) Data-mining models.

4.3.3.1 ANN based models

ANNs are able to capture the autocorrelation structure in a time series even if the underlying law
governing the series is unknown or too complex to describe. Since quantitative forecasting is
based on extracting patterns from observed past events and extrapolating them into the future,
thus ANN may be assumed to be good candidates for this task. The available NN models are:

(i) Multilayer feed forward NN (FFNN)

(ii) Radial basis function network (RBF)
(iii) Support vector machine (SVM)

( 17 )
(iv) Self-organizing map (SOM),
(v) Committee machine of NNs
(vi) Recurrent neural network (RNN).

Neural networks are highly interconnected simple processing units designed in a way to model
how the human brain performs a particular task. Each of those units, also called neurons, forms a
weighted sum of its inputs, to which a constant term called bias is added. This sum is then passed
through a transfer function: linear, sigmoid or hyperbolic tangent.
Fig. 1 shows the internal structure of a neuron. Multilayer perceptrons are the best known and
most widely used kind of neural network. Networks with interconnections that do not form any
loops are called feedforward. Recurrent or non-feedforward networks in which there are one or
more loops of interconnections are used for some kinds of applications. The units are organized
in a way that defines the network architecture. In feedforward networks, units are often arranged
in layers: an input layer, one or more hidden layers and an output layer. The units in each layer
may share the same inputs, but are not connected to each other. Typically, the units in the input
layer serve only for transferring the input pattern to the rest of the network, without any
processing. The information is processed by the units in the hidden and output layers.

( 18 )
 Fig.1 Internal Structure of Neuron

Fig 2. Example of three layered feedforward neural network model with single output

( 19 )
Fig. 2 shows the architecture of a generic three-layered feedforward neural network model. The
neural network considered is fully connected in the sense that every unit belonging to each layer
is connected to every unit belonging to the adjacent layer. In order to find the optimal network
architecture, several combinations were evaluated. These combinations included networks with
different number of hidden layers, different number of units in each layer and different types of
transfer functions. We converged to a configuration consisting of a one hidden layer that uses a
hyperbolic tangent sigmoid transfer function, defined as:
f (s) = 1 es (1)
1 + es
where s is the weighted input of the hidden layer, and f(s) is the output of the hidden layer. The
output layer has only one unit with a pure linear transfer function. This configuration has been
proven to be a universal mapper, provided that the hidden layer has enough units [33]. On the
one hand, if there are too few units, the network will not be flexible enough to model the data
well and, on the other hand, if there are too many units, the network may overfit the data.
Typically, the number of units in the hidden layer is chosen by trial and error, selecting a few
alternatives and then running simulations to find out the one with the best results. Forecasting
with neural networks involves two steps: training and learning. Training of feedforward networks
is normally performed in a supervised manner. One assumes that a training set is available, given
by the historical data, containing both inputs and the corresponding desired outputs, which is
presented to the network. The adequate selection of inputs for neural network training is highly
influential to the success of training. In the learning process a neural network constructs an
inputoutput mapping, adjusting the weights and biases at each iteration based on the
minimization of some error measure between the output produced and the desired output. Thus,
learning entails an optimization process. The error minimization process is repeated until an
acceptable criterion for convergence is reached. The knowledge acquired by the neural network
through the learning process is tested by applying new data that it has never seen before, called
the testing set. The network should be able to generalize and have an accurate output for this
unseen data. It is undesirable to over train the neural network, meaning that the network would
only work well on the training set, and would not generalize well to new data outside the training

( 20 )
set. Overtraining the neural network can seriously deteriorate the forecasting results. Also,
providing the neural network with too much or wrong information can confuse the network and
it can settle on weights that are unable to handle variations of larger magnitude in the input data.
The most common learning algorithm is the back propagation algorithm, in which the input is
passed layer through layer until the final output is calculated, and it is compared to the real
output to find the error. The error is then propagated back to the input adjusting the weights and
biases in each layer. The standard back propagation learning algorithm is a steepest descent
algorithm that minimizes the sum of square errors.
However, the standard back propagation learning algorithm is not efficient numerically and tends
to converge slowly. In order to accelerate the learning process, two parameters of the back
propagation algorithm can be adjusted: the learning rate and the momentum. The learning rate is
the proportion of error gradient by which the weights should be adjusted. Larger values can give
a faster convergence to the minimum but also may produce oscillation around the minimum. The
momentum determines the proportion of the change of past weights that should be used in
the calculation of the new weights. An algorithm that trains a neural network 10100 times faster
than the usual back propagation algorithm is the Levenberg- Marquardt algorithm. While back
propagation is a steepest descent algorithm, the Levenberg-Marquardt algorithm is a variation
of Newtons method. Newtons update for minimizing a function V(x) with respect
to the vector x is given by:
x=[ 2 V ( x )]1 V (x ) (2)
where 2 V ( x ) is the Hessian matrix and V (x ) is the gradient vector.
Assuming that V (x) is the sum of square errors, given by:
N

V(x) = e2h(x) (3)

h=1
then:
V (x )=2 J T (x ) e( x ) (4)
2 V ( x )=2 J T ( x) J (x )+ 2 S(x ) (5)
where e(x) is the error vector,

J(x) is the Jacobian matrix given by:

( 21 )
J(x) =
e1(x) e2(x) . eN(x)
x1 x1 x1

e1(x) e2(x) .. eN(x)

x2 x2 x2

.
. .
. .
e1(x) e2(x) .. eN(x)
xn xn xn

and where S(x) is given by:

N

e h(x)
S(x) = 2
(7)

h=1

Neglecting the second-order derivatives of the error vector, i.e., assuming that S(x)0, the
Hessian matrix is given by:
2 V ( x )=2 J T ( x) J (x ) (8)

and substituting Eqs. (8) and (4) into Eq. (2) we obtain the GaussNewton update, given by:
x=[J T (x)J ( x)]1 J T ( x)e (x) (9)

The advantage of GaussNewton over the standard Newtons method is that it does not require
calculation of second-order derivatives. Nevertheless, the matrix JT(x) J(x) may not be
invertible. This is overcome with the Levenberg-Marquardt algorithm, which consists in finding
the update given by:

( 22 )
 x=[J T (x)J ( x)+ I ]1 J T (x) e ( x) (10)

where parameter is conveniently modified during the algorithm iterations. When is very
small or null the Levenberg-Marquardt algorithm becomes Gauss-Newton, which should provide
faster convergence, while for higher values, when the first term within square brackets of
Eq. (10) is negligible with respect to the second term within square brackets, the algorithm
becomes steepest descent. Hence, the Levenberg-Marquardt algorithm provides a nice
compromise between the speed of Gauss-Newton and the guaranteed convergence of steepest
descent.

4.3.3.2Data-mining models.
Recently, data-mining techniques like Bayesian categorization method, closest k-neighborhood
categorization, reasoning based categorization, genetic algorithm (GA) based categorization,
have gained popularity for data interpretation and inferencing.
Fuzzy model is one of the data mining techniques used for price forecasting.

4.3.3.3.1Fuzzy Model
Fuzzy inference used forecasting purpose is basically a working of fuzzy logic in which rules
which are linguistic are used to map the input onto output space without any firm requirement of
inputs.
Fuzzy Logic:
if X is a universe of discourse with elements denoted by x, then the fuzzy set A in X is defined as
a set of ordered pairs,
A = {x, A(x) |x X}
A(x) is called the membership function of in A.
Fuzzy inference systems (FIS) uses fuzzy rules (IF Then) and fuzzy reasoning which
performs input-output mapping based on fuzzy logic and it is able to handle the concept of

( 23 )
incomplete truth instead of absolute truth. ANFIS (Adaptive network based fuzzy inference
system) is fuzzy inference system which is used for forecasting is basically a FIS which is
implemented in the framework of adaptive network. For utilizing the piece wise continuous
nature of electricity price in the time domain clustering of the input data is done by maintaining
the variations, Thus FIS is used for handling the data that lies beyond the coverage of NN. The
Australian New-South Wales electricity market data was used to test FIS. Using efficient neuro-
fuzzy combination LMPs of the PJM market were calculated in which fuzzy reasoning and RNN
was used. The fuzzy rules were used to perform the linguistic reasoning about the contingencies
and the reasoning results served as a part of inputs to the RNNs in above neuro fuzzy method.
For forecasting day-ahead electricity prices in the Ontario market and PJM market a novel hybrid
intelligent algorithm utilizing a data filtering technique based on wavelet transform (WT), an
optimization technique based on firefly (FF) algorithm, and a soft computing model based on
fuzzy ARTMAP (FA) network was used.

An FIS performs input-output mapping based on fuzzy logic. Fuzzy evaluates the intermediate
states between discrete crisp states and is able to handle the concept of partial truth instead of
absolute truth. Traditional adaptive fuzzy system include ANFIS and neuro-fuzzy methods are
intended to combine the advantages of ANN and fuzzy logic with the difference that ANFIS
architecture has linear output function , whereas neuro-fuzzy systems are essentially a subset of
ANN applied to controls and classification problem .
Wang-mendel suggested an algorithm for implementing FIS for time series prediction and the
same approach was extended to forecast the electricity price. The approach is model free and
heuristic in nature. A common framework called the fuzzy rule base is constructed to combine
both numerical and linguistic information. The numerical information is sampled from
measurements, and the linguistic information interprets the numerical information. The FIS is
able to bridge the gap between interpretability and accuracy by providing a verbally interpretable
rule base and numerical accuracy through training. The FIS using wang-mendel learning
algorithm does not require iterative training making it more efficient than ARMA or GARCH
time series techniques and ANN or neuro-fuzzy intelligent systems. Compared to the black box
nature of Artificial Neural Network (ANN) the Fuzzy Inference System (FIS) provides a

( 24 )
transparent linguistic rule base instead of a black box. The rules may be modified manually to
include expert knowledge. The rule base provides FIS the advantage of interpretability and
transparency. FIS also provides flexibility in choosing predefined membership function. The FIS
algorithm can be modified for higher accuracy and efficiency.

5. FACTORS INFLUENCING ELECTRICITY PRICES

The factors influencing spot prices may be classified on the basis of:
C1 market characteristics,
C2 nonstrategic uncertainties
C3 other stochastic uncertainties
C4 behavior indices
C5 temporal effects.
The different input variables, along with the class they belong to, used by different researchers
are presented in Table 1. There are as many as 40 variables used by different researchers. Most of
the researchers have utilized past experience in selecting the input variables for their respective
model and choice of best input variables for a particular model is still an open area of research.

Class Input variable Time period whose data is used as input

C1 (1) Historical load f(load); (d _ m, t), m = 1, 2, 3, 4, 7, 14, 21, 28

( 25 )
 (2) System load rate,
(3) imports/exports,
(4) capacity excess/shortfall (d, t), (d, t _ 1), (d _ 1, t), (d _ 2, t), (d _ 7, t)
(5) Historical reserves (d, t _ 2), (d, t _ 1), (d, t)
(6) Nuclear,
(7) thermal,
(8) hydro generation,
(9) generation capacity,
(10) net-tie flows,
(11) MRR,
(12)systems binding constraints,
(13) line limits (d, t)
(14) Past MCQ (market-clearing quantity) (d _ 1, t)
C2 (15) Forecast load (d, t _ 2), (d, t _ 1), (d, t)
(16) Forecast reserves,
(17) temperature,
(18) dew point temperature,
(19) weather,
(20) oil price,
(21) gas price,
(22) fuel price (d, t)
C3 (23) Generation outages,
(24) line status,
(25) line contingency information,
(26) congestion index (d, t)
C4 (27) Historical prices f(price); (d _ m, t _ n), m = 0, 1, 2, 3, 4, 5,6, 7, 8, 14, 21,
28, 364 and n = 0, 1, 2, 3, 4.
(28) Demand elasticity,
(29) bidding strategies,
(30) spike existence index,
(31) ID flag (d, t)
C5 (32) Settlement period, (d, t)
(33) day type,
(34) month,

( 26 )
 (35) holiday code,
(36) Xmas code,
(37) clock change,
(38) season,
(39) summer index,
(40) winter index

C1 market characteristics, C2 nonstrategic uncertainties, C3 other stochastic uncertainties,

C4 behavior indices, C5 temporal effects, d day, t settlement period number of the day.

CONCLUSION

The report has presented quite a view of price forecasting and methodologies used for price
forecasting. Various methods with exogenous and non-exogenous variables compute to
sufficiently correct forecasts. Various methods like Regression model, time series model has
been discussed which inculcates the formulation for price computation. Neural networks and
fuzzy neural networks can model non-linear input/output mapping functions. However, the price
of electricity is a time-varying signal and its functional relationships vary rapidly with time.
Electrical energy cannot be effectively stored and the need for stability of the power system
demands a continual balancing of generation and load. On short timescales, most users of
electricity are unaware of, or indifferent to, its price. Transmission bottlenecks usually limit
electricity transportation from one region to another. These factors create the extreme price
volatility, and even price spikes, in the electricity market.

( 27 )
 REFERENCE

1. Electricity price forecasting in deregulated markets: A review and evaluation

Sanjeev Kumar Aggarwal, Lalit Mohan Saini, Ashwani Kumar
2. A neural network based dynamic forecasting model for Trend
Impact Analysis.
Nedaa Agami, Amir Atiya, Mohamed Saleh, Hisham El-Shishiny
3. Mid-term electricity market clearing price forecasting: A hybrid LSSVM and ARMAX
approach.
Xing Yan, Nurul A. Chowdhury
4. Electricity price forecasting using artificial neural networks.
Deepak Singhal, K.S. Swarup
5. Day ahead price forecasting of electricity markets by a mixed data model
and hybrid forecast method.
Nima Amjady , Farshid Keynia
6. Short-term electricity prices forecasting in a competitive market:

( 28 )
A neural network approach.
J.P.S. Catalao, S.J.P.S. Mariano, V.M.F. Mendes, L.A.F.M. Ferreira

( 29 )