07583810 day ahead price forecasting in deregulated electricity market using artificial neural network

Day Ahead Price Forecasting in Deregulated
Electricity Market Using Artificial Neural Network
Ms. Kanchan K. Nargale

Mrs. S. B. Patil

Department of Electrical Engineering,
G. H. Raisoni Institute of Engineering & Technology,
Wagholi, Pune, India.


Department of Electrical Engineering,
G. H. Raisoni Institute of Engineering & Technology,
Wagholi, Pune, India.


Abstract— Now a days the price forecasting plays a very
essential role in a new electricity industry; it helps the
independent generators to set up optimal bidding patterns and
also for designing the physical bilateral contracts. In general,
different market players need to know future electricity prices as
their profitability depends on them. There are many papers have
been presented on the forecasting of electricity market price such
methods are based on time series, artificial intelligence and
hybrid methods. In this paper, the price forecasting is presented
by using feed forward artificial neural network by using
historical price data. Accurately and efficiently forecasting of
electricity price is more important. Therefore in this paper, an
Artificial Neural Network (ANN) model is designed for short
term price forecasting of electricity in the environment of
restructured power market. The proposed ANN model is a fourlayered neural network, which consists of, input layer, two

hidden layers and output layer. Matlab is used for training the
proposed ANN model. Electricity load and wind forecasting can
also be done using this method which helps in planning and
operation of the power system.
Keywords— Electricity Market and Price Forecasting and
Artificial Neural Network (ANN).

I.

INTRODUCTION

In a power market the price of electricity has important for
all activities. But in many countries the electricity industry has
very low competitive energy and has less regulated power. The
price forecasting helps to the different power suppliers to sells
the rational offers in short term. The price forecasting helps to
the electricity industries for the investment decisions and
bidding strategies. It is necessary for estimating the uncertainty
involved in the price. There are many methods are presents till
now for the forecasting of electricity market price. These
methods are based on the artificial intelligence and time series.
In some forecasting of electricity market price uses both
artificial intelligence and time series methods.
The price forecasting plays an important role in electricity
industries; it helps to an independent generator to set the
optimal bidding patterns and physical bilateral contracts [1].
Generally the different electricity industries have needed to
know the future electricity prices and the profitability depends
on them. Electricity price forecasting is very important to study
because the electricity power market and electricity prices are

978-1-4673-9925-8/16/$31.00 ©2016 IEEE

highly volatile in nature. As the degree of volatility of
electricity markets is higher than that of other markets, due to
the risk of volatility is created in every market [2]. Also the
storage of electricity is very costly therefore the electrical
supply and demand needs to be balanced in real time. To
balance the supply and demand properly many numbers of
factors are to be considered such as production of hydro
generation, generating units availability, effects of weather,
changes to prices of related commodities such as fuel price, and
sudden occurred physical problems in transmission systems
and generation. Generally for forecasting purpose there are
different types of forecasting models are used like as traditional
time series models in [2], Auto Regressive Integrated Moving
Average (ARIMA) models, simpler Auto Regressive (AR)
models modern techniques such as ANN, Fuzzy logic [3]have
been used for price forecasting. The traditional price
forecasting models are uses the mathematical model for the
regression analysis and time series analysis. Also there are
many artificial intelligent methods are used for price
forecasting recently. Out of these all methods the ANN method
is very powerful tool and simple for price forecasting. The
ANN method is used in this paper to forecast the price because
this method has high capability to learn the complicated
relationship between the input and output through a supervised
training process with historical data. There are many factors are
affected on the electricity price forecasting, these factors are
line limit, load pattern, bidding pattern and generator outage.

Out of these factors the load pattern is the more effective
parameter for bidding behavior of Generating Companies
(Gencos). Therefore in this paper the historical price and load
patterns are considered to forecast the price. The three layered
feed forward ANN method is used to shows the price
forecasting results. The Historical data for this market is
collected from 2008 to 2010 in 24 hours.
Organization of this paper is in the following way section II
reviews the development of system, the different proposed
methods used in this paper are presents in this section. In
section III the Artificial Neural Network (ANN) Models are
presents. In section IV simulation of the proposed system and
the experimental results are presents. The simulation is done in
MATLAB software. Finally section V concludes this paper.

527


II. DEVELOPMENT OF SYSTEM
This section reports the development of the proposed
method this algorithm has been tested on training data set; also
in this section the different modules are considered for
designing the a good neural network model for short term price
forecasting. Figure 1 indicates the simple flow diagram of the
work done in price forecasting methodology.
A. Collection of Data:
The real time data Market clearing price (MCP) and Market
clearing volume is taken from Indian Energy Exchange, Delhi
(IEX) and Power Exchange India Limited, Mumbai (PXIL)
[4].To find the optimal input parameters ANN uses the input

selection from the collected historical data. By using these
optimal inputs ANN shows the more accurate result and has
great speed. Parameters, which effect on the electricity price
can be categorized into day type (the day of a week), historical
price data and the amount of demand (system load). The
correlation analysis is used to predict the price of previous
hours. In this paper work the data is collected from 2008 to
2010 in 24 hours. Out of this data collection 70% data is used
for training the sequence and remaining 30 % data is used for
validation purpose.
B. Analyzing the Data:
These MCP and MCV are analyzed. From analysis it is
found that price is volatile in nature and this volatility is higher
than any other commodity. The analysis of data is done by
using normalization technique. The formula for the
normalization method is given by,

(1)

Collection of data (MCV &
MCP) from IEX & PXIL
Analysis of data by using
Normalization technique

Developing price forecasting
model using ANN
Price Estimation

Analyzing the Results
Fig.1. Proposed Block Diagram

The normalization technique which is used in this paper has
main advantage of mapping the target output to the nonsaturated sector of tensing function. This technique is useful to
improve the accuracy of both the forecasting modes and
training data sequences.
C. Developing the ANN model for price forecasting
The forecasting model will be developed using artificial
intelligence tool. And this model will trained using the
analyzed data.
D. Training and Validation:
The training process of ANN requires a proper network
inputs and target outputs for forecasting the prices. The training
process the set of examples of data is given to the ANN
network. In the training process the biases and the weights of
an ANN are properly adjusted to minimize the performance of
network function. In this method, historical price data has been
used for forecasting the price in day ahead market.
Out of the total data collected, 70% of the data is used for
training the sequence and remaining 30 % data is used for
validation purpose.
E. Result Analysis:
Result will be analyzed by comparing the actual results and
predicted results, the model can be tested for its efficient
prediction.
III. THE ARTIFICIAL NEURAL NETWORK (ANN) MODELS
The Artificial Neural Network (ANN) based models is the
first technique which is used for the price prediction. This
method has the most popular tool for different price load
forecasting applications.
A. The SVM Models

The Support Vector Machine (SVM) models are one of the
latest techniques which are used for electricity price
forecasting. Most of the recent techniques cannot handle the
nonlinear price forecasting problems properly. But in case of
SVM, it shows the better performance than these methods. The
SVM model uses the statistical learning of theory which is used
to minimize the structural risk, instead of the usual empirical
risk of forecasting and for this purpose it minimizes the upper
bound generalization error. The SVM models are also used for
solving problems of small sample size, classification, and
regression and time series predictions of forecasting.
B. Use of ANN for Price Estimation
The artificial neural network (ANN) is a mathematical
model which is based on biological neural network. The
artificial neural network (ANN) refers to the inter–connections
between the neurons in different layers of each system. This
system has three layers, the first layer consists of input neurons
which are useful to send data through synapses to the second
layer of neurons, and after that more synapses to the third layer
of the output neurons. The ANN contains a group of artificial
neurons which are interconnected to each other’s. The ANN is

528


a nonlinear mathematical modeling tool which can be used for
the complicated relationships between the inputs and outputs.
The figure 2 shows the basic diagram of Artificial Neural
Network (ANN).
C. Single layer feed forward networks:

The neurons are organized in the layers in the layered
neural network. The layered neural network is the simplest
form of neural network in which the input layer of source
nodes are projects on to the output layer of neurons and vice
versa and this process is called the feed forward network. As
shown in figure 2 there are four input layer nodes and four
hidden layer nodes are presents for both input and output
layers. And such network is called the single layered network
[6]-[8].
D. Multilayer feed forward networks:
The second class of a feed forward neural network
distinguishes itself by the presence of one or more hidden
layers, whose computation nodes are correspondingly called
hidden neurons or hidden units. The hidden neurons presents in
the network creates the communication between the external
input and the network output in some useful manner. If the one
or more hidden layers are added to the network, then the
network is able to extract higher –order statistics. Due to the
extra dimension of neural network and extra set of connections,
the network acquires a global perspective despite its local
connectivity.
The main ability of hidden neuron is to extract the higher
order mathematics from the input layer. Figure 3 shows the
feed forward network with one hidden layer and one output
neuron.
The source node presents at the input in the figure are used
to get the input signals which are applied to the neurons. In the
second stage the hidden neurons are used to establish the
communication between the input and the output. The outputs
obtained from this hidden layer are acts as input to the third

layer and rest of the network [9]. In the third layer that is the
output layer the network constitutes the overall response of the
network as shown in figure 3.

Fig.3.TheFeed Forward Network With One Hidden Layer and
One Output Neuron

E. Development of ANN for Price Estimation
• Collection of Data: Prices of January 2009 to April
2010 are collected.
• Selection of Input & Output: 24 MCP of each hour on a
day before and 24 MCP of each hour of same day
before a week give 48 MCP as an input to neural
network. And 24 MCP will be in output layer.
• Total Data: Total samples utilized for training and
validation purpose are 455
• Training and Validation Data: Samples from January
2009 to March 2010 are used. To train the data of
network 70% sample and for test the data 30% samples
are used. April 2010 month sample is used for
revaluation of network by comparing estimated price to
actual prices.
A program is written in Matlab software for training of
neural network.
IV. EXPERIMENTAL RESULTS

Fig.2. Basic Diagram of Artificial Neural Network.

This section shows the experimental results of proposed
Price Forecasting in Day Ahead Market by using Artificial

Neural Network (ANN).
The training data is to be taken firstly. The training is started
with one hidden node and it is increased one by one increasing
the performance of network. At 218 Epochs the performance
goal is met. Total time required to train the network was 32
Minutes. The optimum structure of neural network is 48-1024. Figure 4 shows performance of neural network during
training.
The training data is further proceeding to the ANN model to
estimate the price. Figure 5 shows ANN model for trained

529


prices estimation. By using this model future prices are
estimated for unknown samples.
The samples which are not used in the training or testing are
then evaluated for the performance of neural network. Hourly
prices are predicted for six days: 13th of April to 18th of April
2010 and then compared with the actual prices of these
samples.
To evaluate the performance of an Artificial Neural Network
(ANN) module, we compare its estimated price with those
actual prices. The Mean Absolute Percentage Error (MAPE)
and Absolute percentage error (APE) is calculated using
following formula.
Let Pa be the actual price and Pf be the forecast price. Then,
Absolute percentage error (APE) is defined as,

(2)
And MAPE is given by,

(3)
where N= time block

Fig.4. Performance Graph.

Fig.5. ANN model obtained after Training

Fig. 6.Actual and Estimated Price for 13th April 2010.

The comparison between actual and calculated price values
for 13th April 2010 is shown in figure 6. From this figure it is
observed that minimum Absolute Percentage Error (APE)
0.820077 and maximum Absolute Percentage Error (APE) is
31.89093. The mean APE (MAPE) is 18.6161.
Comparison between actual and estimated price values for
14th April 2010 shown in figure 8. It is observed that minimum
Absolute Percentage Error (APE) is 1.047389 and maximum
Absolute Percentage Error (APE) is 33.36022. The mean APE
(MAPE) is 14.35968.

Fig. 8.Actual and Estimated Price for 14th April 2010.

Comparison between actual and estimated price values for
15th April 2010 shown in figure 9. It is observed that minimum
Absolute Percentage Error (APE) is 1.040385 and maximum
Absolute Percentage Error (APE) is 34.3546. The mean APE
(MAPE) is 15.74521.
Comparison between actual and estimated price values for
16th April 2010 shown in figure 10. It is observed that
minimum Absolute Percentage Error (APE) is 0.626772 and

maximum Absolute Percentage Error (APE) is 31.14988. The
mean APE (MAPE) is 15.64559.

530


Fig.7. Performance plot

Fig. 9.Actual and Estimated Price for 15th April 2010

Fig. 10. Actual and Estimated Price for 16th April 2010.

Fig. 11. Actual and Estimated Price for 17th April 2010.

Fig. 12. Actual and Estimated Price for 18th April 2010.

531


Comparison between actual and estimated price values for
17th April 2010 shown in figure 11. It is observed that
minimum percentage error (APE) is 0.421196 and maximum
APE is 36.98921. The mean APE (MAPE) is 14.351833728.
Comparison between actual and estimated price values for
18th April 2010 shown in figure 12. It is observed that
minimum percentage error (APE) is 3.256485 and maximum
APE is 33.360 33.89013. The mean APE (MAPE) is 17.6309.
V. CONCLUSION AND FUTURE SCOPE
The conclusion of the proposed system is based on the
results obtained from the proposed model. The experimental

results show the reasonably good forecast results. These results
are taken when there is no much fluctuation between each hour
and days. This work is an attempt to the study and analyses the
market prices in day ahead market with reference to Indian
electricity market. The data is available on market clearing
prices (MCP) Indian Energy Exchange (IEX) and Power
Exchange India Limited (PXIL). The Artificial Neural network
(ANN) is developed using last two years data to predict hourly
market price. Results obtained from neural network model are
satisfactory.
The tool used for developing this proposed work is ANN in
Matlab software. It is used for training the proposed ANN
model. The performance of the forecasting model can be
improved by considering the various parameters affecting the
price volatility and also by using more historical price data
which will indicate the behavior of price volatility in more
detail.
This work will be further improved to increase the efficiency
in forecasting the electricity price in day ahead market using
the support vector machine tool in Matlab.

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]
[16]

ACKNOWLEDGMENT
We thank the Department of Electrical Engineering, G.H.
Raisoni Institute of Engineering and Technology, Savitribai
Phule Pune University, Pune, Maharashtra, India for
permitting us to use the different computational facilities for
this research and development work.

[17]

[18]

[19]

REFERENCES
[1]

[2]

[3]

[4]
[5]

“An ARIMA Approach to Forecasting Electricity Price with Accuracy
Improvement by Predicted Errors”, Ming Zhou, Student Member, IEEE,
Zheng Yan, Yixin Ni, Senior Member, IEEE and Gengyin Li, Member,
IEEE.
“Several-Hours-Ahead Electricity Price and Load Forecasting Using
Neural Networks”, ParasMandal, TomonobuSenjyu, Member, IEEE,
Katsumi Uezato, and Toshihisa Funabashi, Senior Member, IEEE.
“Electricity price forecasting with confidence-interval estimation
through an extended ARIMA approach” M. Zhou, Z. Yan, Y.X. Ni, G.
Li and Y. Nie.
“Electricity Price Forecasting based on GARCH Model in Deregulated
Market”, ZHENG Hua, XIE Li, ZHANG Li-zi.
“Day-Ahead Electricity Price Forecasting in a Grid Environment”,
Guang Li, Student Member, IEEE, Chen-Ching Liu, Fellow, IEEE,

[20]

[21]

[22]

Chris Mattson, and Jacques Lawarrée, IEEE TRANSACTIONS ON
POWER SYSTEMS, VOL. 22, NO. 1, FEBRUARY 2007.
Jun Hua Zhao, Zhao Yang Dong, Member IEEE, Xue Li, Member IEEE
and Kit Po Wong, Fellow IEEE, “A General Method for Electricity

Market Price Spike Analysis” 2005 IEEE.
Wei Sun, Jian-Chang Lu, Ming Meng, Department of Economics &
Management, North China Electric Power University, Baoding, Hebei
071003, China E-MAIL: , “Application Of
Time Series Based SVM Model On Next-Day Electricity Price
Forecasting Under Deregulated Power Market”, Proceedings of the Fifth
International Conference on Machine Learning and Cybernetics, Dalian,
13-16 August 2006.
S. Fan, C. Mao and L. Chen “Next-Day Electricity-Price Forecasting
Using A Hybrid Network” IET Gener. Transm. Distrib., Vol. 1, No. 1,
January 2007.
Jun Hua Zhao, Student Member IEEE, Zhao Yang Dong, Senior
Member, IEEE, Xue Li, Member, IEEE, and Kit Po Wong, Fellow,
IEEE “A Framework for Electricity Price Spike Analysis With
Advanced Data Mining Methods” IEEE Transactions On Power
Systems, Vol. 22, No. 1, February 2007.
J. Bastin, J. Zhu, V. Banunarayanan, R. Mukerji" Forecasting energy
prices in a competitive market", IEEE Computer Applications in Power,
July 1999.
B. R. Szkuta, L. A. Sanabria and T. S. Dillon, "Electricity Price ShortTerm Forecasting using Artificial Neural Networks", IEEE Transactions
on Power System, Vol. 14, no. 3, pp. 851-857, 1999.
Wei Sun, Jie Zhang School of Business and Administration, North
China Electric Power University, Baoding, 071003, P.R. China
, 2008 International Conference on Internet
Computing in Science and Engineering.
Yi Wang, School of Business Management, North China Electric Power
University, Beijing, 102206, China, Songqing Yu2, Shandong Linqing
Power Supply Company, Linqing, 252600, China, “Price Forecasting by
ICA-SVM in the Competitive Electricity Market” 2008 IEEE.
Jianhua Zhang Jian Han Rui Wang GuolianHou, “Day-ahead Electricity

Price Forecasting Based on Rolling Time Series and Least SquareSupport Vector Machine Model”, IEEE 2011.
Modelling Power Plants Emission Constraints, WSEAS Transactions on
Systems, Vol.3, Issue 7, 2004, pp. 2512-2516.
V. Mendes, J.C. Quadrado, J. Catalão, S. Mariano and L. Ferreira,
Environmental Concerns on Fossil Fuel Plants Emission for Power
Engineering Education, WSEAS Transactions on Advances in
Engineering Educations, Vol.1, Issue 1, 2004, pp. 1-5.
F.J. Nogales, J. Contreras, A.J. Conejo and R. Espínola, Forecasting
Next-Day Electricity Prices by Time Series Models, IEEE Transactions
on Power Systems, Vol.17, No.2, 2002, pp. 342-348.
J. Contreras, R. Espínola, F.J. Nogales and A.J. Conejo, ARIMA Models
to Predict Next-Day Electricity Prices, IEEE Transactions on Power
Systems, Vol.18, No.3, 2003, pp. 1014-1020.
O.B. Fosso, A. Gjelsvik, A. Haugstad, B. Mo and I. Wangensteen,
Generation Scheduling in a Deregulated System, IEEE Transactions on
Power Systems, Vol.14, No.1, 1999, pp. 75-81.
A.J. Wang and B. Ramsay, A Neural Network Based Estimator for
Electricity Spot-Pricing with Particular Reference to Weekend and
Public Holidays, Neurocomputing, Vol.23, Issue 1, 1998, pp. 47-57.
B.R. Szkuta, L.A. Sanabria and T.S. Dillon, Electricity Price Short-Term
Forecasting Using Artificial Neural Networks, IEEE Transactions on
Power Systems, Vol.14, No.3, 1999, pp. 851-857.
F.Gao, X. Guan, X.-R. Cao and A. Papalexopoulos, Forecasting Power
Market Clearing Price and Quantity Using a Neural Network Method, in
Proc. IEEE Power Engineering Society Summer Meeting, Seattle, WA,
2000, pp. 2183-2188.

532