Electricity Demand Forecasting Using Fuzzy-Neural Network

Electricity Demand Forecasting Using Fuzzy-Neural
Network
Naren Chandra Kattla
Department of Computer Science Engineering
VIT University, Chennai-600128, India
Phone: +91-9765434698
Email: kattlanaren.chandra2015@vit.ac.in
Abstract—Demand forecasts are important for energy sup-
pliers, electricity generation, transmission, distributors. Neuro-
fuzzy systems have attracted growing interest of researchers in
various scientific and engineering areas due to the increasing need
of intelligent systems. The methods considered are fuzzy neural
network (FNN) and an artificial neural network (ANN) trained
using backpropagation (BP) algorithm. The implementation and
forecasting hybrid fuzzy neural technique, which combines neural
network modelling, and techniques from fuzzy logic and fuzzy set
theory for electric load forecasting. The strengths of this powerful
technique lie in its ability to forecast accurately.
I. INTRODUCTION
Expanding access to electricity means including the rural
areas people of 1.2 billion and still 0.6 billion there is no
access to electricity and 0.4 billion people there is access to
unreliable electricity networks. We need machine technique
and practical approaches, as a mechanism that receives inputs
directly and transmits to development, electricity plays a major
role in both fighting against poverty and climate change.
Forecasting using Artificial neural network(ANN) and Fuzzy
Neural Network(FNN) had achieved satisfactory results.
The implementation and comparing different techniques are
followed by
• Artificial Neural Network
• Draw Back Of Neural Network
• The Proposed Approach
• Fuzzification
• Neural Network
• Defuzzification
II. ARTIFICIAL NEURAL NETWORK
A. Data Encoding And Normalization
One of the essential keys to working with artificial NN is
data encoding and normalization.
After all data has been collected we may have some data
in non-numeric data and high magnitude data. After all data
has been converted to Encoding non-numeric output data to
numeric values by using 1-of-N encoding, encoding non-
numeric input data to numeric values by using 1-of-(N-1)
encoding and apply normalization to high magnitude using
any one out of two different types of normalization, Gaussian
normalization and min-max normalization.
In general, the min-max normalized value for some value
x is
(x − min)
(max − min)
(1)
very simple, the Gaussian normalized value for some value x
is
(x − mean)
Standard Deviation
(2)
B. Neural Network
An artificial neural network will accept one or more
inputs and produces one or more outputs. The basic neural
network input to process (hidden layer) and process to output
computation is known the feed-forward NN mechanism.
Fig. 1: Neural Network Architecture
Understanding the feed-forward NN mechanism is impor-
tant to know how to create NN, which will make predictions.
A NN computes in several stages to get output and the feed-
forward NN mechanism is explained by using (Fig. 1), there
are six inputs each line connecting one node to another repre-
sents a weight constant. The weight is labelled IHWEIGHT [0]
[0] it represents the weight from input GDP to hidden 1 and
the weight in the lower right corner is labelled HOWEIGHT

[12] [0] it represents the weight from hidden 7 to output. All
hidden and output layer nodes has an arrow pointing to it this
are called the bias values.
Fig. 2: Activation Functions
The first stage in the feed-forward NN mechanism is to
compute for all hidden nodes. The value of hidden 1 node
and is computed as the product of each input value and its
associated weight are summed then associated bias value is
added. From (Fig. 2) Pre-activation sum function applied by
the activation function, it will be explained in detail next
section, but for now it’s enough to say that the activation
function is the hyperbolic tangent function, which is usually
abbreviated tanh.
Hidden Node 1 = tanh(Pre Activation Sum) (3)
After all, hidden node output values have been computed,
these values are used as inputs to the output nodes and it’s
computed slightly differently from the hidden nodes. The pre-
liminary output sums, before activation, for output nodes are
computed the same way as hidden node sums, the activation
function for the output layer is called the softmax function.
For 1st Output Node:
output =
eP re Activation Output Sum
eP re Activation Output Sum
(4)
For nth Output Node:
output =
a
b
(5)
Form equation 5, a = eP re Activation Output Sum
and b =
eP re Activation Output 1 Sum
+ ... + eP re Activation Output n Sum
If we look at (Fig. 2), we will see these are the final output
values computed by the neural network. Where exp function
is the math constant e = 2.71828. . . raised to the xth power.
Notice the output values sum is 1.0, which is not a coincidence
that is the point of using the softmax function.
1) Activation Functions: In (Fig. 2) for hidden layer node
activation function is hyperbolic tangent and for the output
layer node activation functions is softmax function are used
and there is a third common activation function called the
logistic sigmoid function. In general, the hyperbolic tangent
function is the best choice for hidden layer activation. For
output layer activation, if neural network is performing classi-
fication where the dependent variable to be predicted has three
or more values softmax activation is the best choice. If neu-
ral network is performing classification where the dependent
variable has exactly two possible values the logistic sigmoid
activation function is the best choice for output layer activation.
C. Training
The ultimate goal of a neural network is to make a
prediction. In order to make a prediction, a neural network
must first be trained. Training a neural network means finding
a set of good weights and bias values so that the known outputs
of some training data match the outputs computed using the
weights and bias values. The resulting weights and bias values
for a particular problem are often collectively called a model
and it’s used to predict the output for unseen inputs that is
computed for known output values.
The data set consists of a total of 5 items data set and
splits it randomly into a 4-item subset (80%) to be used for
training and a 1-item subset (20%) to be used for testing,
that is, to be used to estimate the probability of a correct
classification on data that has not been seen before. Training
a 6-7-1 fully connected feed-forward NN using the back-
propagation algorithm in conjunction with a technique that is
called incremental training.
After training has completed, the accuracy of the resulting
model is computed on the training set and on the test set.
III. DRAW BACKS OF USING NEURAL NETWORK
These ANN based methods had given better performance
when compared to traditional techniques, the accuracy is
limited due to the following limitations in ANN:
• The accuracy is depended on the training data set, So
selecting correct training data to cover the entire poses
problems is difficult.
• The neural networks need to re-training due to change
in seasons.
These methods lack when compared to common-sense
knowledge frequently used by human experts to forecast load
profiles. The predictive accuracy of forecasting models de-
pends on projecting values and interrelationships of important
parameters (weather).
In order to overcome problems and to improve the fore-
casting, a unique approach by combining the powers of neural
network and fuzzy logic techniques. This hybrid techniques is
good by combining both features to overcome the limitations
of each.

IV. THE PROPOSED ARCHITECTURE
The main idea behind this hybrid approach is : the fuzzy
knowledge-base models the knowledge about the system and
its input parameters, quantitative as well as qualitative, and the
neural network captures the inexplicable relationship between
fuzzy inputs and outputs.
Fig. 3: Flow of processing
Form (fig. 3)inputs to this hybrid model include the daily
temperature, rain forecast, seasons, day type. A fuzzy front-end
processor has been developed for preprocessing these inputs
with the application of fuzzy rules. The front-end processor
effectively handles the different types of inputs, both numeric
as well as fuzzy, and produces a fuzzy output vector which is
then fed to a three- layer back propagation neural network.
During training, the neural network captures the unknown
mapping between these input variables and the target outputs.
Once trained, the outputs of the neural network, interpreted as
fuzzy membership functions of the target load, are defuzzified
to obtain the load profile for the following day.
V. FUZZIFICATION
A. Fuzzification variables
The inputs to the fuzzy consist of the daily temperature,
rain forecast, seasons, day type. The first step consists of
defining all inputs and outputs as fuzzy sets and variables
values are defined range of possible values for each variable
(Fig. 4).
Fig. 4: Fuzzification of input
The output variables is electricity consumed.
B. Membership Function
A membership function (MF) is a curve that defines how
each point in the input space is mapped to a membership value
(or degree of membership) between 0 and 1. For the following
variable the membership function as show below:
1) Temperature Membership Function: The daily temper-
ature for the day are represented range is [25 45] and these
are divided into three linguistic classes each [low (L), medium
(M), high (H)] as in the (Fig. 5).
Fig. 5: Membership function of Temperature
2) Rain Forecast Membership Function: The rain forecast
has been found to play a dominant role in the accuracy of the
forecast.The weather forecast in terms of rain is represented
as fuzzy subsets [very heavy (VH), heavy (H), medium (M),
light showers (LH), no rain (NR)] (Fig. 6).
Fig. 6: Membership function of Rain Forecast
3) Seasons Membership Function: In seasons the range
is [1 4]and these are divided into four linguistic classes for
spring[March, April, May], summer[June, July, August], au-
tumn[September, October, November], winter[December, Jan-
uary, February] (Fig. 7).
4) Day Type Membership Function: The distance of a
weekdays (or) weekends (for example, from a Saturday or
Monday to Sunday) the range is [0 4] and it is divided into
four segments[war (F), near (N), very near (VN), day (D] (Fig.
8).
5) Electricity Consumed Membership Function: The
change in load is represented by a fuzzy variable P, divided into
five referential sets [negative big (NB), negative small (NS),
zero (ZE), positive small (PS), positive big (PB)], as shown in
(Fig. 9).

Fig. 7: Membership function of Seasons
Fig. 8: Membership function of Day Type
Fig. 9: Membership function of Electricity consumed
C. Identiﬁcation of Day
In graph on x-axis is weekly cycles and on y-axis is
electricity consumed load curve, with random variations as
seasons changed the electricity consumed load is increas-
ing(winter, spring, summer). Based on the load curve, day type
are grouped in to two separate classes:
1) Weekday (Monday to Friday): The shape of the load
curve on all weekdays, Friday is high when compare to other
days (Fig. 10)
2) Saturday and Sunday: The load-curve on Saturdays,
most businesses are open for the ﬁrst half of the day, is
Fig. 10: Typical load shape for weekdays
substantially different from the rest of the days,due to this
high load demand in the morning.The load curve on Sundays,
where less when compare to Saturday (Fig. 11). By this load
on holidays is very less compare to businesses days.
Fig. 11: Load on weekends
D. Fuzzy Rules
In most case, more than one fuzzy rules are applied for
same output variables. For example, the fuzzy rules:
Rule 1 : IF Temperature is High THEN Load is High
Rule 2 : IF Rain is V ery Heavy THEN Load is Low
E. Training Patterns
The training patterns in the learning of the fuzzy-neural
network consist of Sugeno-Type Fuzzy Inference System(FIS)
is used in this input variables applied with membership values
of p and fuzzy inference rules. The outcome of fuzzy rules
as described above (Fig. 12). The Target load is output of
electricity consumed and it also load into FIS with membership
function (Fig. 9). This processed data combines different kinds
of knowledge before being fed to the neural network.

Fig. 12: Fuzzy rule based system
Fig. 13: preparation of Traning Set
VI. NEURAL NETWORK
The role of the neural network is to capture the unknown
mapping between fuzzy input and output vectors. Form (Fig.
14) First layer is associated with each input variables, Second
layer for each input membership function is determined by in-
putmf, Third layer is rules by the fuzzy truth table proposition,
In Fourth layer for each Sugeno-Type Fuzzy Inference System
the output membership value will be generated and Fifth layer
is defuzzification for generated outputmf.
The 3-layer neural networks able to capture the electric-
ity consumed for Weekdays and weekends. The activation
function for hidden and output is Sigmoid transfer. Back-
propagation type algorithm, which calculates the error rates,
defined as the derivative of the squared error with respect to
each node’s output, recursively from the output backward to
the input nodes is used for learning.
The number of neurons for representing these variables is
5, 3, 4, 4 and 240 respectively. The training process is shown
in (Fig. 15).
Fig. 14: Adaptive Neuro-Fuzzy Inference System
Fig. 15: Neural Network Training
VII. DEFUZZIFICATION
This is the final level, at this point output value from the
neural network is a fuzzified in the range [0 - 1]and it will
indicates the degree of membership function of the outcome
in the range [Xmin -Xmax]. This result is defuzzified by
converting it in this pre-specified range to obtain the daily
load value (in MW) for each week of seasons.
X = Xmin + Xn ∗ (Xmax − Xmin) (6)
where Xn is the predicted output of the NN and X is the
corresponding load value in MW.
VIII. RESULTS
This fuzzy neural network has been implemented using
Matlab (Fig. 16). The knowledge base and the membership
functions along with 240 rules were train by 10 epochs and
Error tolerance 0.By repeatedly Fine training from observa-
tions (Fig. 17) hybrid system giving better results by using
climate conditions.

Fig. 16: Neuro-Fuzzy Designer
Fig. 17: Training Result
IX. CONCLUSIONS
A forecasting technique based on ANN and FNN results
form this hybrid technique approach can be used to elec-
tricity load forecasting with greater accuracy than the ANN
technique. The FIS can be easily developed and modified to
reflect changes in different seasons. The basic concepts of this
hybrid technique(FNN) are clearly presented in this paper and
to implement on a personal computer and allows for operator
intervention is very simple. The results obtained based on
weather parameters are found to be very accurate and superior
when compared to the ANN approach.
REFERENCES
[1] A.Kumar Singh, Ibraheem, S. Khatoon, Md. Muazzam, ”An Overview
of Electricity Demand Forecasting Techniques, National Conference
on Emerging Trends in Electrical, Instrumentation and Communication
Engineering, vol. 3, no.3, 2013.
[2] R. Xu, G. K. Venayamoorthy, and D. C. Wunsch, Modeling of gene
regulatory networks with hybrid differential evolution and particle swarm
optimization, Neural Networks, Vol. 20, pp. 917 927, 2007.
[3] Azadeh, S. F. Ghaderi, S. Tarverdian, and M. Saberi, Integration of
artificial neural networks and genetic algorithm to predict electrical energy
consumption, Applied Mathematics and Computation, Vol. 186, pp.
17311741, 2007.
[4] F. Hobbs, S. Jitprapaikulsarn, and S. Konda, Analysis of the value for
unit commitment of improved load forecasts, IEEE Trans on Power
Systems, Vol. 14, no. 4, pp. 13421348, 1999.
[5] S. A. Billings, and X. Hong, Dual-orthogonal radial basis function
networks for nonlinear time series prediction, Neural Networks, Vol.
18, pp. 479493, 1998.
[6] L. Yue, Y. Zhang, Q. Zhong, and Z. Wang, Mid-long term load forecast-
ing based on fuzzy logic clustering neural network approach, Journal
of North China Electric Power University, Vol. 35, no. 2, pp. 4246, 2008.
[7] A. Kazemi et al., A Multi-Level Artificial Neural Network for Gasoline
Demand Forecasting of Iran, In Second International Conference on
Computer and Electrical Engineering, pp. 61-64, 2009.
[8] A. Ghanhari, A. Naghavi, S.F. Ghaderi and M. Sabaghian, Artificial
Neural Networks and regression approaches comparison for forecasting
Irans annual electricity load, Proceeding International Conference on
Power Engineering Energy and Electrical Drives. pp. 675-679, 2009.
[9] Sackdara, et al., Electricity Demand Forecasting of Electricity DU Lao
(EDL) using Neural Networks, TENCON 2010, pp. 640-644, 2010.

Electricity Demand Forecasting Using Fuzzy-Neural Network

More Related Content

What's hot (19)

Viewers also liked (6)

Similar to Electricity Demand Forecasting Using Fuzzy-Neural Network (20)

Electricity Demand Forecasting Using Fuzzy-Neural Network