1

Electricity Price Forecasting Techniques

Akhilesh kumar, 2014JEN2594

AbstractElectricity price forecasting has become an essential

tool since the deregulation of electricity market. Expected price
profiles can help market participants to determine their optimal
bidding strategies and, consequently, maximize their benefits.
This paper presents a variety of methods or techniques of
Electricity Price Forecasting that has been tried in various
markets, their comparative studies and to figure out best among
them.

that affect the supply and demand balance and hence play
a critical role on price volatility. Such factors include hydro
generation, availability of generating units, sudden changes in
weather conditions, changes to prices of related commodities
such as fuel price, and unexpected physical problems in
generation and transmission systems [1]. The main factors
influencing electricity price are shown in figure 1.

Index TermsElectricity Price Forecasting, Electricity market,ANN, Time series

I. I NTRODUCTION
NE of the main reasons why electricity price forecasting is an important study is the very volatile nature
of the electricity market. In the deregulated market, where
customer is a king and he has right to choose among different
electricity suppliers which makes it extremely important to
know the supply-demand balance ahead of time for all market
players and particularly for generating companies and hence
the electricity price, must be pre-estimated before real time
operation for maximizing profit. Price forecasting plays a
key role in the electricity industry like helping independent
generators in setting up optimal bidding patterns, designing
physical bilateral contracts, and also gives price signals for
the decision on investing a new generation facilities in the
long run. In general, different market players need to know
future electricity prices as their profitability depends on them;
whether it is the generating companies or the ISO, large
industrial customers or investors, it is very critical to have this
forecast. Thus the objective of Price Forecasting model is to
visualize future electricity market dynamics to assist in present
decision making process. Load and demand forecasting tools
have been widely developed and deployed globally but usage
in India is still low. There is immense need for such a tool but
complex requirements in India may not be met by a standard
product or model. It needs to consider dynamics of Indian
power sector, demand patterns, weather patterns and various
other local data inputs; which makes it a unique solution.This
has arises the need to precisely predict future electricity prices,
a hot issue in the research area. This paper reviews established
approaches to electricity price forecasting and indicates the
potential future development of methodologies for accurate
electricity price forecasting.

II. FACTORS

AFFECTING ELECTRICITY PRICE

The degree of volatility is higher on electricity markets

than other markets. The main reason being electrical supply
and demand need to be balanced in real time since it is not
economical to store electricity.There are a number of factors
Department of Electrical Engineering, IIT Delhi, India

Figure 1. Influencing factors of electricity price

There are a number of parameters that affect the bidding

strategy of the generating companies. There are technical
constraints on unit operation; load and weather forecast and
hydro energy availability are some of the factors. The variables
would vary with regard to the forecasting dimensions, with
different input variables for Long and medium term forecast
which would not vary considerably in shorter time frame while
on the other hand the short term forecast or Day ahead, hour
ahead prices forecast the input variables would vary as per the
market characteristics. In long term forecasting hydro factor
is crucial as the water level in the reservoir is high would
help to moderate the electricity prices in the market. The long
term Power purchase agreements (PPAs) are also required to
be taken into consideration as this would have implication of
prices. The volatility in the prices can be largely attributed
to the nature of electricity which is different from a normal
commodity as it cannot be stored and there is need to maintain
balance between demand and supply, also the inelastic nature
of demand over short period of time. These factors needs to
be categorized into input variable and these variables could
be as many 50 nos put across different buckets to understand
the impact of the each variable on the final price output by
studying the past behavior.
III. D IFFERENT M ETHODOLOGY FOR PRICE

FORECASTING

The price forecasting can be applicable in various time

frames based on their specific requirements but for short term

forecast its accuracy can be low due to the incomplete information or uncertain bidding strategy of the market participants but
the accuracy is more reasonable with the time frame enlarged.
The purpose of classifying the forecasting in different time
frames is due to the fact that there is a different forecasting
methods and approach which is applicable for each of the
time frames. It is not advisable to use the method of short term
forecast for long term forecast or otherwise because short term
forecast uses quantitative approach and long term forecast uses
strategic and business decisive approach. Different methods in
Forecasting are categorized as mentioned below:
Qualitative methods - where there is no formal mathematical model (long-term forecasting)
Regression methods Here a variable is thought to be linearly related to a number of other independent variables
Multiple equation methods There are number of dependent variables that interact with each other through a
series of equations (as in economic models)
Time series methods There is a single variable that
changes with time and whose future values are related in
some way to its past values. These techniques are mostly
used for short and medium term forecasts.
There are various initial activities to be performed before the
actual forecast is initiated. Basic activities to be performed
are:
Data Management Large Volume of real time and
Historic data from various sources need to be collected
and stored in a common database.
Data Preparation Processing Configuration Validation
Analysis and Estimation of data
Build Data Model
Forecasting and Optimization Model
A first step to price forecasting is to identify all the crucial
factors and variables which set the future price trend in the
market. The initial step is the data preparation and doing
data analysis thereby clearly identifying the odd variables
so that the forecasting result is not biased based on them.
The historic models and its results need to be studied to
select the most efficient model. A basic model is developed
initially which can be clearly understood by the power market
players and then a comparison between the actual and forecast
results is done to do the required optimization of model.
There are various method used for electricity price forecasting
and it becomes difficult to identify the appropriate method
for forecasting. To understand the price forecasting methods
better it is important to understand the influencing factors and
forecast these influencing factors using the existing and proven
methodologies like in the case of weather forecasting or be the
load and demand forecasting tools. But these tools may not
be very effective and errors would vary for example Artificial
Intelligence (AI) approaches such as neural network, and
support vector machine (SVM), which have been successfully
applied in load forecast whereas Support Vector Regression
(SVR) model which is developed in terms of the particularity
of the price forecast in electricity market.
It is also important that the input variables used for the
Electricity price forecasting are also predicted using proven

methodologies and required modeling is done for them. Figure

2 illustrates one of such approach whereby different input
variables are predicted such as the weather input parameters
like temperature humidity etc. is predicted using the weather
forecast module and on the other Load forecasting module is
utilized for the prediction of load parameters.

Figure 2. Price Forecasting Module

For the purpose of Long and medium term forecasting it

would be required to setup basic database on the installed
capacity and the additional projects in the pipeline for power
generation which would be input for Supply forecast. The next
step is to input the required data along with historical datas
for all the influencing factors also the historic price data. The
availability of accurate historic data is a key factor for accurate
forecast. The screening of data for accuracy and also in case
of some data is missing that needs to be estimated. The data
analysis is to be carried which shall be followed by selection of
an appropriate model. The next step is to do the application
development which shall integrate various models and also
do the required reporting. The process is to be repeated with
different models to arrive at the optimized state wherein the
right model combination is selected for each module and this
requires various permutation and combinations to be carried
out with due consideration to time factor. This is an iterative
process, for an investor who would like to know the Return
on Investment (ROI) for his investment and the best method is
to calculate the net present value of additions in resources or
withdrawals for each year of the forecast period is considered.
It is predicted to see of the future prices of electricity can
be sufficient to cover the resource, development, operation,
maintenance costs this includes his profit margin.
There is another approach for short and medium term forecasting which is detailed market simulation approach in which
lot of data pertaining to power exchange market is required.
This is preferred by the Power utilities and market operators.
In this method the actual market condition is simulated by
assuming the demand and supply data based on the historic
trend and also factor for system operating constraints.
This method can be used as supporting tool over and
above the statistical and intelligent models. These approaches
forecast future prices using historical operation data. The need

for different modules is to accurately predict the Demand

Supply ratio as it is observed that the electricity prices follow
the demand supply curve and hence the demand supply ratio
is major contributor as variable input for the price prediction
model.

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. Study of game
theory models in itself is a major area of research and has been
kept outside the scope of this paper. A detailed discussion on
game theory models can be found in [4].
B. Simulation model

Figure 3. Demand Supply Ratio

IV.

ELECTRICITY PRICE FORECASTING MODEL

Many techniques or models have been evolved over

the years but still this area is an area of research.
There are two categories of forecasting methods, parametric methods and artificial intelligence based methods.
Price forecasting models can be classified into three sets
these three sets have been further divided into subsets which are shown in figure 4. A new branch of
data mining methodologies has been added to the tree.

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 [5] and the other is UPLAN
software developed by LCG Consulting[6].
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. Where as 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, fuel prices
and demand forecasts.
Simulation methods are proposed 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.
C. Time series model

Figure 4. Classification of forecasting methods

A. Game theory models

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.

A time series is a set or sequence of observed data arranged

in chronological order and in an equally spaced time intervals
such as daily or hourly air temperature. Time series analysis is
a method of forecasting which focuses on the past behavior of
the dependent variable. Sometimes external variables can also
be included within a time series framework. Based on time
series, there are further three types of models.
1) Parsimonious stochastic models: Stochastic models are
taken from the financial literature and a desire to adapt some
of the well known and widely applied in practice approaches.
These are discrete time counterparts which correspond to

continuous time stochastic models. Some discrete type models

are autoregressive (AR), moving average (MA), autoregressive
moving average (ARMA), autoregressive integrated moving
average (ARIMA) and generalized autoregressive conditional
heteroskedastic (GARCH) have been considered.
Stochastic time series can be divided into stationary process
and non-stationary process. The basic assumption of stationary
process 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. As electricity price is a non-stationary process, which exhibits daily, weekly, yearly and other periodicity.
Therefore, a different class of models that have this property,
designated as seasonal process model, is used
Figure 5. Example of ANN model

2) Regression or casual : 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 external variables. The
explanatory variables of this model are identified on he basis
of correlation analysis on each of these independent variables
with the price (dependent) variable.

3) Artificial intelligence (AI) models: These may be considered as non-parametric 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 and (ii) data-mining models.

4) ANN based models.: Artificial Neural Network (ANN)

based models have been found to be the most popular for
load forecasting applications. Like wise, ANNs appeared to be
among the first techniques to emerge in the price prediction
concept as well.
To recall the basics, ANN works in the same way as a
human brain does. It is composed of highly interconnected
neurons working in unison to accomplish a specific task.
Neural networks learn by example and relate a set of input
variables xi ,i=1,2,3,...n to a set of output variables yi ,i =
1,2,3..n . The typical structure of an ANN model is shown in
Fig.5. During the training process, neurons in the input layer
pass the raw information onto the rest of neurons in the other
layers, without any processing. The weights between neurons
keep updating according to supervised learning. Based on the
measures of minimal error between the output produced and
the desired output, the process is repeated until an acceptable
error is reached. This training process is called back propagation. After the model acquires the knowledge, new data can
be tested for forecasting.

The ANN models could be different with regard to combinations of different numbers of hidden layers, different numbers
of units in each layer and different types of transfer function.
Generally three layers of neural networks are chosen but in
some papers four layers has been used for price forecasting.
Neural networks have well-known advantages of being able
to solve undefined relationships between input and output
variables, approximate complex nonlinear functions and implement multiple training algorithms. However, neural networks
also have the following disadvantages: the network will not
be flexible enough to model the data well with too few units,
and on the contrary, it will be over-fitting with too many units.
The wavelet and ANN models are fitted together for greater
price forecasting accuracy .
D. Data mining
Data inference and interpretation is done by several recently
found techniques like genetic algorithm (GA) based categorization method, Bayesian categorization method, closest kneighborhood based categorization method, reasoning based
categorization method etc.
V.

DIFFERENT METHODS USED FOR FORECASTING

A. ARIMA ( Auto Regressive Integrated Moving Average

Model)
This Model has been successfully applied for the load and
price forecasting. Nowadays, it is also being used for prediction of short term electricity prices. ARMA, the combination
of auto-regressive (AR) and moving-average (MA) models, is
defined by Pt =t +1 Pt1 +1 t1 +t
Where Pt is electricity price at time t; t is the mean of
price at t;1 is auto-regressive coefficient of the price series;1
is moving-average coefficient; t is a stochastic value with
expectation zero and square error 2 . ARIMA models can
be reduced to ARMA models through the pre-processing
of the electricity price series. After the pre-processing, the
price series becomes a stationary time-series such that ARMA
models can then be applied. In particular, the ARIMA methods
are extended to include error correction for the worse market

conditions with high price volatility [8]. In [9], techniques

that based on the wavelet transform and ARIMA models are
applied to Spanish power markets in order to improve the
accuracy of price forecasting.
B. Hybrid Models
the mean time, different integrated techniques have also
been proposed by many researchers. An efficient tool for one
step ahead forecasting was created and merged with some
other methods of prediction. They have been compared and
tested for long time. Single models have been compared
with Multivariate models and results prove that accuracy of
forecasting got improved. Some proposed examples are:

Bayesian Clustering Dynamics (BCD) & SVM.

Clustering & Wavelet analysis applied on networks.
VI. ACCURACY C RITERIA

To measure the forecast accuracy, we choose from

several indexes, such as Mean Absolute Error(MAE)
and
Mean
Absolute
Percentage
Error(MAPE):

Where Yt the real data at time stage t;Yt is the forecast

output data at t.
MAPE is frequently used to analyze the error of the
load forecasting results, since it is more robust and easy to
understand. However, the electricity price could be zero or
even negative depending upon the bidding behavior in spot
power markets.
VII. C ONCLUSIONS
The Electricity price forecasting is complex but based on
the various research and case studies it is essential tool which
needs to be applied with thorough analysis as there is no
standard model or scheme which can adopted off the shelf
but approach is to have a different models combined together
come out with an expert model and can used alternatively for
different forecast period.
Based on a survey of established approaches to electricity
price forecasting, this paper gives an over view of price
forecasting, with a summary of the influencing factors of price
behavior and extended taxonomy of price forecasting methods.
Different forecasting methods,such as ANN, ARIMA models
and are reviewed separately. The hybrid methods that combine
different models to offset the weakness of individually established models are also highlighted in order to clearly indicate
the future trend of electricity price forecasting methodology.

R EFERENCES
[1] Hamidreza Zareipour, Price Forecasting and Optimal Operation of
Wholesale Customers in a Competitive Electricity Market, Ph.D. dissertation, Dept. Electrical and Computer Eng., Univ. of Waterloo, Canada,
2006
[2] Electricity Price Forecasting Model - Defining the Need and Approach
for India Market Jasdev Singh Soni BE (E&E), PG (Power Plant
Engineering), PGDBA (Operations Management) International Journal of
Enhanced Research in Science Technology & Engineering, ISSN: 23197463
[3] Byounghee Kim, John P. Velas, Jeongkyu Lee, Jongbae Park, Joongrin
Shin, and KwangY. Lee, "Short-Term System Marginal Price Forecasting
Using System-Type Neural Network Architecture,"Power Systems Conference and Exposition,2006.
[4] Bajpai, P, Singh SN. Bidding and gaming in electricity market: an
overview and key issues. In: Proceedings of national power system
conference (NPSC), Chennai; 2004. p. 338346
[5] Bastian J, Zhu J, Banunaryanan V, Mukherji R. Forecasting energy prices
in a competitive market. IEEE Comput Appl Power 1999(July):405.
[6] Deb R, Albert R, Hsue Lie-Long, Brown N. How to incorporate volatility
and risk in electricity price forecasting. Electricity J 2000(May):116
[7] Moghram I, Rahman S. Analysis and evaluation of five short-term load
forecasting techniques. IEEE Trans Power Syst 1989;4(4):148491.
[8] M. Zhou, Z. Yan, Y.X. Ni, G. Li and Y. Nie, "ElectricityPrice Forecasting with Confidence-interval Estimation through an Extended ARIMA
Approach,"lEE Proc.-Gener. Tansm. Distrib.,Vol.153,No.2,March 2006.
[9] Antonio1. Conejo, Miguel A. Plazas, Rosa Espinola,and Ana B.
Molina,"Day-Ahead Electricity Price Forecasting Using the WaveletTransformand ARIMAModels,"IEEE Transactions on Power Systems,Vo1.20, No.2, May 2005.