IJEAS0704008

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

International Journal of Engineering and Applied Sciences (IJEAS)

ISSN: 2394-3661, Volume-7, Issue-4, April 2020

Short Term Load Forecasting Using Artificial Neural


Network & Time Series Methods
Suman Adhikari, Prof. Dr. Laxman Poudel

Abstract- Short term Electric load forecasting is an important efficiency scores. Load forecasts can have significant
aspect of power system planning and operation for utility implications on energy transactions, market shares and profits
companies. Short term load forecasting (STLF) has always been in competitive energy markets. Forecasting electricity prices
and quantities in daily power markets are the most essential
one of the most critical, sensitive and accuracy demanding factors
task and basis for any decision making.
of the power systems. An accurate STLF improves not only the
systems economic viability but also its safety, stability and The short term forecasts refers to hourly prediction of the load
reliability. The researcher presented in this works support for a lead time ranging from one hour to several days out. The
mid-term forecasts can either be hourly or peak load forecasts
Artificial Neural Network and Time Series Methods techniques in
for a forecast horizon of one to several months ahead. Finally
short term forecasting. This paper presents an investigation for
the long term forecasts refer to forecasts made for one to
the short term (one day to seven days, & every months of one several years in the future.
year) load forecasting the load demand of Nepal Electricity
Authority (NEA) in Bishnumati Feeder of Balaju Substation, by
A power system serves a major function of supplying its
customers, (large and small) with economical and reliable
using artificial neural network and time series methods.
electrical energy as much as possible. For adequate electricity
to be supplied to the customers, their load demand must be
Keywords – Short Term Load Forecasting (STLF), Neural known. The process of making these evaluations of future
Network, Backpropagation algorithm, Moving average demand of load is called ‘Load forecasting’. Load forecasting
is the projection of electrical load that will be required by a
certain geographical area considering previous electrical load
1. INTRODUCTION usage in the said area Load forecasting helps to make vital
decisions concerning the system, therefore, load forecasting is
Electric power is the one of the most major input factors in very crucial for successful effective and efficient operation of
economic development. To support economic growth and any energy system. If the system load forecast is
meet power requirements continually in the future, load overestimated, the system may overcommit the generation of
forecasting has become a very important task for electric power which will inadvertently lead to costly operation of the
utilities. Moreover, an accurate load forecast can helpful in power system. On the other hand, if the system load forecast is
developing a power supply strategy, financing, planning, underestimated, the reliability and security of the system may
market research and electricity management. However, up to be compromised, resulting in power interruptions and customer
now, the general style of load forecasting emphasized dissatisfaction. The time period in which the forecast is carried
aggregate load forecasting. Such load forecasting results not out is fundamental to the results and use of the forecast. Short-
only cannot identify where the power load takes place but also term forecast, which spans a period of one hour to one week,
is not helpful for power facilities construction location helps to provide a great saving potential for economic and
planning. Besides, on the one hand, people want to enjoy the secured operation of power system, medium-term forecast,
convenience of electricity in their daily life, but on the other which ranges from a week to a year, concerns with scheduling
hand, no one wants to have any power facilities constructed in of fuel supply and maintenance operation and long-term
their backyard. Regional load forecasting can determine the forecast and is from a year upwards, is useful for planning
electricity power shortage area directly. On the basis of operations.
equity, power facilities may be constructed in the power
shortage region. The accuracy of forecast greatly depends upon the size of the
power system network, reliability and outages. Forecasting
Load forecasting, at several levels of power networks became becomes more complex with the smaller system. Hence, more
a key instrument in modern power system operation. The accurate methods and additional information is required to get
reason for its growing importance is related, on the one hand acceptable outputs. It becomes more complex with
to the emergence of the energy markets, demanding higher unpredictable occurrence of anomalous events. The quality and

11 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

quantity of available load data, metrological data, sociological


data etc. directly affect the quality of result. Identification of
proper tool, management of information system has another
important role in the model development. Good forecasting
with limited information and with large number of constraints
is major challenge faced by the all the power system engineers
all around the world. The challenges is more pronounced by
the fact that the information and constraints are completely
different from one power system to other or one part of the
world to other. No general algorithm or tool can be developed.
Hence, for each system a specific method and algorithm has to
be developed regarding information collection, data
management, data filtration, model development and scope of
implementation.

There are large varieties of mathematical methods that are used


for load forecasting, the development and improvements of
suitable mathematical tools will lead to the development of
more accurate load forecasting techniques. The accuracy of
load forecasting depends on the load forecasting techniques
used as well as on the accuracy of forecasted weather
parameters such as temperature, humidity etc. As per the
recent trends artificial intelligence methods are the most
pronounced for the STLF. From different artificial intelligence Figure 1: Forecasting Methodology by Artificial Neural
methods, fuzzy logic and artificial neural network are the most Network (ANN)
used. Among the two methods fuzzy logic for STLF is gaining.
The load forecasting methodology by moving averages and
exponential smoothing technique can be shown in flow chart:
II. METHODOLOGY

This study is focused on the short term load forecasting


techniques using the different model of short term load
forecasting. Similar day approach to the modern artificial
intelligence has been studied and implemented all over the
world in various power grid best suit. In Nepal, short term load
forecasting is not implemented in real hour to hour or day to
day forecasting because no accurate model and technique
cannot be implemented in power sector of Nepal.

In this research different analytical technique are analyzed on


the basis of Mean Absolute Percentage Error (MAPE) and
Mean Absolute Deviation (MAD) between the actual and
forecasted load. Both techniques use the feed forward back
propagation neural network algorithm for both data training
and data forecasting.

The basic Framework for this research methodology are listed


is shown below

Figure 2: Load Forecasting Model for Moving Average &


Exponential Smoothing

12 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

From these two method of short term load forecasting along with temperature variables, day type & cyclic values of
technique by artificial neural network & time series method hour of day and day of week as input.
(moving averages & exponential smoothing) the error can be
calculated. The least efficient error method be the most With properly synchronized additional variables the
appropriate method for the load forecasting. Therefore the performance error MAPE, MSE etc. gets improved
ANN method be the best method of the load forecasting automatically.
because these method having minimum error on the
forecasting. Thus the formulated input vector and target variable is feed in
the ANN to get predicted values using 2- layer feed forward
2.1. Tool Selection network with sigmoid hidden neurons and linear output
neurons. It can fit multi-dimensional mapping problems
For, all the predictions related to ANN Matlab 2018a is used abnormally well given consistent data and enough neurons in
and for time series methods statistical tool excel is used. its hidden layer. The neural network is trained with Levenberg-
Marquardt back propagation algorithm.
2.2. Evaluation of Predicted Performance
The forecasting performance of the trained network could be
assessed by calculating the prediction error on samples other
than those used during the training phase. Various error
metrics between the actual and forecasted loads are presented
and defined but the most commonly adopted by load
forecasters are the Mean Absolute Percentage Errors (MAPE),
the Absolute Percentage Errors (APE), the Mean Absolute
Error (MAE) and the Mean Squared Error (MSE) or Root
Mean Squared Error (RMSE).

Fig 3: Neural Network Training Regression

Training of the Network: This is just a process of weight


adjustments with respect to the targeted output by the neural
network. The training of the network is carried out using the
MATLAB software. The load data collected is the input data to
the neural networks. 50% of this data was used for the training
Where n is the number of the data points and i is the period at of the neural network, 25% was used for the validation and the
which the load is produced or forecasted, t is the target and O remaining 25%, for the forecast. The neural network was
the NN output. trained using different activation functions and number of
To make sure that the system is accurate, the relative error is layers till the best performance was obtained. Figure 5 shows
retained on the hourly basis. In the case of positive error, it the neural network being trained.
means the forecasted load is greater than the actual
consumption load, and the opposite is true when the forecasted 2.4. Data Collection
load was less than the actual load. Combined hourly feeder data from 1 st Poush 2075 to 29th
Poush 2076 of the Bishnumati feeder of 132/11 Kv, collected
2.3. Forecasting next hour using ANN from Balaju Substation, Nepal Electricity Authority. The raw
data obtained had been recorded manually in Balaju Substation
Two input methods are tested for forecasting next hour load is
containing single excel sheet for daily load diagrams. The
set that the effects of power variables and additional input
obtained data was then converted into single time series data.
parameters. The forecast is done with historical load variables

13 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

Similarly, hourly metrological data (temperature, humidity) Table 2: Input and Target Parameter for ANN Model
was collected from department of Meteorology, Government
of Nepal. Ho Previou Temperatur Humid worki Target(
ur s e(Degree ity(%) ng MW)
In this method of load forecasting using ANN method, we can Load(M centrigrate) hour/
be split the data into the inputs set, hidden layer sets, output W) not
layer and the output. For, this load forecasting using ANN 1 0.97 4 88 0 0.97
methods we can be use temperature, previous hour load, 2 0.97 5 87 0 0.97
humidity, in which days of load be forecast & whether the day 3 0.97 6 88 0 0.97
be working day or not. The overall parameter of the neural 4 0.97 6 86 0 0.97
network can be listed in table.
5 0.94 6 89 0 1.13
Table 1: Neural Network Architecture 6 1.10 7 84 0 1.67
7 1.68 8 88 1 2.30
Number Number Number Activation 8 2.32 9 89 0 2.45
of of Hidden of Output Functions
9 2.46 9.5 85 0 2.25
Inputs layer Neuron
Neurons 10 2.20 10 67 0 2.20
5 10 1 Sigmoid 11 2.19 12 70 0 2.17
12 2.07 17 73 0 2.14
13 2.09 18.5 75 0 2.14
2.5. Time Series Architecture
14 2.17 22 76 1 2.07
In this method of load forecasting we can be used two methods
15 2.06 19 77 0 2.14
of time series analysis. These are moving average and the
exponential smoothing technique. In this method of load 16 2.04 16 76 0 2.11
forecasting the moving of three item can be analyzed and then 17 2.02 14 74 0 2.54
be take into the fourth column. The exponential smoothing 18 2.36 13 76 0 2.96
technique can also be used the same technique, at the starting 19 2.91 12 74 0 2.79
we can be take the averages of hourly load and forecasted the 20 2.75 11 75 0 2.43
load for next hour, next day and the next month.
21 2.38 6 74.5 0 2.11
3. Objectives 22 2.04 5 74.6 1 1.52
23 1.49 4 74.2 0 1.20
3.1 Main Objective
24 1.17 4 74.5 0 1.04
The main objective of this thesis is short term load forecasting
using Artificial Neural Network and Time Series Method.
The load Forecasting can be done on the on the following
3.2. Specific Objectives basics. The input and target value are given to the matlab
toolbox and then analyzing the regression line. If the
 To find the hourly basis data of temperature, humidity
regression line is very close to unity, then there could be the
& power. accuracy of forecasted load.
 To identify the major network disturbances.
Table 3: Calculation of Forecasted load & APE
 To validate the data result with actual data.
Ho Actual load, Forecasted Absolute
4. Results & Discussion ur MW load, MW Percentage Error
1 0.94 0.974823 0.037859
4.1. Forecasted load with ANN Model
2 0.94 0.975202 0.038262
The hourly load data of the Bishnumati Feeder 132/11 can be 3 0.91 0.974341 0.074394
collected from the Balaju Substation. First of all we can set the 4 0.91 0.971354 0.0711
input and the target model for the load forecasting. The input
5 1.13 0.993563 0.123528
parameters are day, previous load, temperature, humidity & the
hour is working hour or not. The input & targeted parameter of 6 1.75 1.32966 0.239749
the 2076 mangsir 1st are as follows: 7 2.30 1.68225 0.268452

14 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

8 2.48 2.468294 0.003801 The mangsir 1st load can be analyzed on the basis of Moving
9 2.28 2.283354 1.23E-05 point averages:
10 2.27 2.563162 0.130547 Table 4: Load forecasting using Moving Averages of mangsir
11 2.23 2.403972 0.075699 1st
12 2.17 2.126062 0.020258
3Point MA 4 Point 5 Point
13 2.04 2.042409 0.000951 MA MA
14 2.01 2.714401 0.351739 Time(h Actual(M Forecast( Forecast( Forecast(
15 1.88 1.880662 0.001137 rs.) W) MW) MW) MW)
16 2.01 1.883253 0.062163 1 0.97 1.57 2.17 2.49
17 2.43 2.577683 0.061155 2 0.97 1.21 1.77 2.32
18 3.00 2.839645 0.052165 3 0.97 1.04 1.42 1.97
19 2.79 2.766072 0.00694 4 0.97 0.97 1.15 1.61
20 2.48 2.469414 0.003349 5 1.13 0.97 0.97 1.33
21 2.04 2.22909 0.09244 6 1.67 1.03 1.01 1.00
22 1.46 1.466409 0.006128 7 2.30 1.26 1.19 1.14
23 1.17 1.129685 0.031131 8 2.45 1.70 1.52 1.41
24 1.04 1.02559 0.010458 9 2.25 2.14 1.89 1.70
10 2.20 2.33 2.17 1.96
11 2.17 2.30 2.30 2.17
%MAPE = 6.34%
12 2.14 2.21 2.27 2.27
The graph that can be compares the Actual Load and 13 2.14 2.17 2.19 2.24
Forecasted load of the Mangsir 1st load, the maximum load in 9 14 2.07 2.15 2.16 2.18
hour and minimum in 1st hour.This gives actual forecast:
15 2.14 2.12 2.13 2.14
16 2.11 2.12 2.12 2.13
17 2.54 2.11 2.11 2.12
18 2.96 2.26 2.21 2.20
19 2.79 2.54 2.44 2.36
20 2.43 2.76 2.60 2.51
21 2.11 2.73 2.68 2.57
22 1.52 2.44 2.57 2.57
23 1.20 2.02 2.21 2.36
24 1.04 1.61 1.81 2.01
%MAP 12% 11.23% 11.44%
E
The MAPE of these forecasted load be slightly higher than the
ANN method again then other 5 days of one week MAPE are
listed in table below.

Figure 4: Actual Vs Forecasted Load of mangsir 1st Table 5: Forecasted Value for different day of Mangsir

4.2. Forecasted load on the basis of Moving Average Moving Average MAPE%
Date 3 Point 4 Point 5 Point
Load can be forecasted on the basis of moving average and the
exponential smoothing. This type of method uses the load at Mangsir 2nd 10.34% 10.55% 11%
input and the output can be analyzed on the basis of 3 point, 4 Mangsir 3rd 9% 9.04% 10.35%
point, 5 point moving averages. Exponential smoothing Mangsir 4th 10% 9.55% 11%
technique is very efficient and beneficial method. Mangsir 5th 11% 10.45% 12%
Mangsir 6th 10.5% 10.50% 11%

15 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

Mangsir 7th 9.5% 11% 10.42%

4.3. Forecasted load on the basis of Exponential Smoothing


Exponential Smoothing is the best technique for electrical load
forecasting. The forecasting of load for seven days of Mangsir
be listed as follows. The forecasted load can be calculated on
the basis of different value of alpha. The forecasted load for
the seven days can be calculated as follows & all the
forecasted load are in Mega Watt.

Table 6: Forecasted load on the basis of Exponential


Smoothing
Figure 5: Comparison between Actual load and Forecasted
a=0.1 a=0.2 a=0.3 load at alpha 0.1.
Time(Hr.) Actual Forecast Forecast Forecast When the value of alpha be 0.3, the graph that can be plotted
load load load load as:
1 0.972 0.972 0.972 0.972
2 0.972 0.972 0.972 0.972
3 0.972 0.972 0.972 0.972
4 0.972 0.972 0.972 0.972
5 1.134 0.972 0.972 0.972
6 1.668 0.988 1.004 1.520
7 2.300 1.056 1.140 2.215
8 2.445 1.180 1.389 2.540
9 2.251 1.307 1.642 1.812
10 2.202 1.401 1.831 1.943
11 2.170 1.481 1.991 2.021
12 2.138 1.550 2.129 2.066
13 2.138 1.609 2.246 2.087
14 2.073 1.662 2.352 2.102
15 2.138 1.703 2.434 2.094 Figure 6: Forecasting using exponential smoothing using alpha
16 2.105 1.746 2.521 2.107 0.3
17 2.542 1.782 2.593 2.106
Others day of the mangsir are listed are as follows:
18 2.964 1.858 2.745 2.237
19 2.785 1.969 2.966 2.455 Table 7: Forecasted load by exponential smoothing
20 2.429 2.051 3.129 2.554
Forecasted MAPE%
21 2.105 2.088 3.205 2.517
0.1 0.2 0.3
22 1.522 2.090 3.208 2.393 st
Mangsir 1 8% 9% 8.46%
23 1.198 1.233 3.095 1.432
mangsir 2nd 7.46% 7.99% 8.20%
24 1.036 1.250 1.928 1.852
mangsir 3 6.22% 7% 7.79%
%MAPE 8% 9% 8.46%
The graph that can be plotted below in figure: mangsir 4 6.89% 8% 8%
mangsir 5 6.42% 7% 8.22%
mangsir 6 7.02% 7.40% 8.11%
mangsir 7 7.49% 8.42% 8.5%

16 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

This can be indicates that MAPE is maximum on the mangsir 23 1.12968 2.02 2.033279 1.20
1st load and indicates minimum in the Tuesday of the normal 5
week. Therefore the mape is maximum on Sunday, because of 24 1.02559 1.61 1.949788 1.04
Sunday be the starting period of the week, but the Tuesday be
normal days of week. Therefore mape is maximum in starting
and minimum in Tuesday of a week. Other day are fluctuating 4.4. Comparison of MAPE for different days.
in-between these maximum and minimum ranges.
The electric load can be forecasted on the basis of next hour
Da ANN Moving Exponential Actual forecast and the next day forecast. The MAPE error of the next
y Average Smoothing Load hour error is less than the next day error, because in next day
1 0.97482 1.57 0.972 0.97 error there should neglected different parameters of input.
3
2 0.97520 1.21 0.971965 0.97 MAPE for different days
2 Next hour Next Day
3 0.97434 1.04 0.971934 0.97 M
1 ovi Expon Mov
A
4 0.97135 0.97 0.971906 0.97 ng ential AN ing Exponential
Day N
4 Av smoot N Aver smoothing
N
era hing agee
5 0.99356 0.97 0.97188 1.13 ge
3
6. 7.
6 1.32966 1.03 0.988052 1.67 12 12.2
Sun 34 8% 34 9%
7 1.68225 1.26 1.056047 2.30 % 4%
% %
8 2.46829 1.70 1.1804 2.45 10
4 4. 6.
Mo .3 7.46 11.3
9 2.28335 2.14 1.306892 2.25 76 67 7.55%
n 4 % 2%
4 % %
%
10 2.56316 2.33 1.401302 2.20 3. 5.
2 9 6.22 9.40
Tue 67 43 7%
11 2.40397 2.30 1.481413 2.17 % % %
% %
2
3. 6.
12 2.12606 2.21 1.550274 2.14 We 10 6.89 10.7
78 34 7.22%
2 d % % 2%
% %
13 2.04240 2.17 1.60901 2.14
4. 6.
9 11 6.42 11.2
Thu 66 46 6.66%
14 2.71440 2.15 1.661873 2.07 % % 2%
% %
1
15 1.88066 2.12 1.702971 2.14 10
5.
2 .5 7.02 7 10
Fri 84 7.42%
16 1.88325 2.12 1.746437 2.11 0 % % %
%
3 %
17 2.57768 2.11 1.782318 2.54 5. 9. 6.
7.49
3 Sat 56 50 86 9% 8%
%
18 2.83964 2.26 1.858335 2.96 % % %
5
19 2.76607 2.54 1.968856 2.79 From these three methods of load forecasting using next day
2 forecasting using hourly basis. The ann method have a
20 2.46941 2.76 2.05051 2.43 minimum error as compared with the moving average and the
4 exponential smoothing method. The maximum error occurred
21 2.22909 2.73 2.088372 2.11 in Sunday because of peak day and minimum error in Tuesday.
22 1.46640 2.44 2.09006 1.52 The comparison graph as shown in below:
9

17 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

Table 4.12: %MAPE for different month

Month ANN Moving Average Exponential


Smoothing
Poush- 4.76% 7% 5%
1
Magh- 6.74% 9% 8%
1
falgun- 5.76% 9.20% 6%
1
chaitra 6.76% 10% 10%
-1
Baisakh 7.44% 9% 8%
-1
Jestha- 8.95% 12% 11%
1
Asadh- 6% 12% 12%
1
Figure 7: Next hour MAPE of different days of week Shawa 6% 10% 8%
n-1
Again, the load forecasting can be done rather than the average Bhadra 6.50% 10.92% 7%
of hour and can be used directly previous days load -1
Ashwin 6.67% 10.62% 6%
-1
Kartik- 6% 8% 8%
1
Mangsi 5.52% 8.72% 6.42%
r-1

From these table, we can be conclude that the maximum


MAPE occurs in jestha month and the minimum MAPE on
poush. Because of monsoon change in jestha ashar month and
load will be fluctuated, then there will be maximum error
occur. The histogram plot of MAPE of different method are
shown bwlow

Figure .8: Comparison of MAPE for next day

4.3. Calculation of MAPE for different Month

The calculation of MAPE for different month can be calculated


on the basis of input and the target parameters. The input
parameter be previous days load, temperature, humidity,
whether the working day be worked or holiday, and the
number of days that can be used. The forecasted load can be
calculated on the basis of regression line. The MAPE for the
different month can be listed as follows.

Figure 9: %MAPE comparison of different month

18 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-7, Issue-4, April 2020

These shows the maximum and minimum value of MAPE for 2, pp. 194–202, 2019, doi: 10.30534/ijatcse/2019/14822019.
different month can be listed in above figure.
[8] J. Sruthi and R. L. H. Catherine, “A Review on Electrical Load
Forecasting in Energy Management 1 1,” vol. 2, no. 3, pp. 670–676,
5. Conclusion 2015.

Artificial Neural Network model was developed incorporating [9] Y. Yang, Y. Meng, Y. Xia, Y. Lu, and H. Yu, “An Efficient
the statistical method. In this research we can be compared the Approach for Short Term Load Forecasting,” vol. I, 2011.
statistical method and the artificial method, the mean absolute
percentage error of the ANN be less than as compare with the
moving average and the exponential smoothing technique. Suman Adhikari: Dept. of Mechanical Engineering Institute of Engineerng,
Pulchowk Campus, TU, Pulchowk, Lalitpur
An artificial neural network method can be used different
parameters as an input while the statistical method can only Prof. Dr. Laxman Poudel: Dept. of Mechanical Engineering Institute of
Engineering, Pulchowk Campus, TU, Pulchowk, Lalitpur
use one input. Therefore the accuracy of the ANN method of
load forecasting can be more accuracy than the other methods.

The MAPE of the mangsir 1st be 6.34% and 12% for ANN
method and the moving averages method. From these we can
be concluding that load forecasting can be done on the basis of
neural network.

ACKNOWLEDGEMENT

The authors extend their hearty gratitude to the faculties and


colleagues at Tribhuvan University, Institute of Engineering,
Pulchowk Campus and engineers of Nepal Electricity
Authority for their constant help, support and
recommendations.

REFERENCES

[1] C. C. Hsu and C. Y. Chen, “Regional load forecasting in Taiwan -


Applications of artificial neural networks,” Energy Convers. Manag.,
vol. 44, no. 12, pp. 1941–1949, 2003, doi: 10.1016/S0196-
8904(02)00225-X.

[2] M. D., A. A. Ayeni, and M. A., “Short Term Electric Load


Forecasting using Neural Network and Genetic Algorithm,” Int. J.
Appl. Inf. Syst., vol. 10, no. 4, pp. 22–28, 2016, doi:
10.5120/ijais2016451490.

[3] S. S. Reddy and J. A. Momoh, “Short term electrical load forecasting


using back propagation neural networks,” 2014 North Am. Power
Symp. NAPS 2014, pp. 1–6, 2014, doi:
10.1109/NAPS.2014.6965453.

[4] S. Saravanan, S. Kannan, and C. Thangaraj, “INDIA ’ S


ELECTRICITY DEMAND FORECAST USING REGRESSION
ANALYSIS AND ARTIFICIAL NEURAL NETWORKS BASED
ON PRINCIPAL COMPONENTS,” vol. 6956, no. July 2012, pp.
365–370, 2020, doi: 10.21917/ijsc.2012.0057.

[5] A. K. Jain, J. Mao, and K. M. Mohiuddin, “Artificial neural


networks: A tutorial,” Computer, vol. 29, no. 3. pp. 31–44, 1996,
doi: 10.1109/2.485891.

[6] M. Buhari and S. S. Adamu, “Short-Term Load Forecasting Using


Artificial Neural Network,” no. March, pp. 1–7, 2012, doi:
10.13140/2.1.3687.0085.

[7] M. K. Singla, J. Gupta, and P. Nijhawan, “Comparative study on


backpropagation and levenberg marquardt algorithm on short term
load forecasting,” Int. J. Adv. Trends Comput. Sci. Eng., vol. 8, no.

19 www.ijeas.org

You might also like