Article Review
Article Review
Article Review
Submitted by:
Jonathan C. Pacaldo
MEP-EE
Submitted to:
Dr. Allan Soriano
Professorial Lecturer
ISSUES TO ADDRESS:
Main Question:
What is the best method to forecast Electrical Load?
Specific Question:
What are the available methods used in forecasting electrical load?
Paper search history:
Search Tool: Science Direct search from ISI Web
Year Span: 2010-2014
Keywords
No. Of Articles
Forecasting
Load Forecasting
Electrical Load Forecasting
Electrical Load Forecasting Methods
30, 838
7, 031
3, 050
2, 804
PAPER 2
Title:
Author/s:
Publication:
1071
PAPER 3
Title:
LONG
TERM
LOAD
FORECASTING
FOR
THE
EGYPTIAN NETWORK USING ANN AND REGRESSION
MODELS.
Author/s:
Publication:
Research Comparison:
Study
Paper 1
Paper 2
Paper 3
Objectives
Study the long term peak load forecasting for the city of
Kutahya, Turkey.
To predict 1-day ahead electrical power load under the
influence of e using a combination of Wavelet and Elman
networks as a recurrent neural network.
Long Term load forecasting for Egyptian unified network
and comparing the result generated by ANN (Artificial
Neural Network) Model and Regression Model.
Studied Methods
Paper 1
2.
3.
4.
5.
6.
Paper 2
1. Daubechies Wavelet
2. Elman Network
Paper 3
Paper 1
literature.
Paper 2
Paper 3
Where,
n number of data
yi, real value existing recorded data
yi, approximate type of the function used
S number of the squared prediction errors
In this method, the variable S is equalised to zero after differentiated to each
coefficient.
A. SIMPLE LINEAR REGRESSION METHODS
The simple linear regression model is based on the linear relationship
between the dependent variable y and independent variable x as shown in equation
(2).
This is the equation of a straight line which intercept y axes at a with a slope
of b. In order to attain zero error, by using the least squares errors method,
equations (3) and (4) are obtained.
In the equations the variables y, x and n represent the peak load, the years
and the number of years which the forecasting is based on respectively. The a and
b coefficients are calculated from the Equations (3) and (4) and replaced in equation
(2) for the load forecasting.
B. MULTIPLE LINEAR REGRESSION
This approach shows a plane in the space with three dimensions which can
be expressed as given in equation (5).
By writing the equation (9) in logarithmic form and then applying the least
squares approach, equations (10), (11) and (12) are formed.
Since the Equation (10) is linear, by applying linear trend analysis a and b
coefficients are formed as shown in equations (13) and (14).
By replacing a and b coefficients in Equation (9) the peak loads are predicted.
ARTIFICIAL NEURAL NETWORKS
Figure 1 depicts the architecture of typical feed-forward multilayered neural
network consist of an input layer, (one or more) hidden layers and an output layer.
The number of hidden layers and neurons in layers is subject to problem studied
and decided upon trial-error.
The input layer receives the signal from outer environment and distributes it
to the neurons in the hidden layers. The hidden layers have computational neurons
and the number of layers depends on the functions to be used. The network
computes the actual outputs of the neurons in the hidden and output layer by using
the activation function. The error gradient for the neurons in the output layer is
calculated and the weights in the back-propagation network propagating backward
the errors associated with output neurons are adjusted. The total error at the output
layer is then reduced by redistributing the error backwards through the hidden
layers until the hidden layer is reached. The process updating the weights until the
desired output is reached defined as training. This process is called as generalised
delta rule and repeated until the error criterion for all data sets is reached. In
general, each ANN is trained on a different 80% of the training data and then
validated on the remaining 20%. Since each additional layer exponentially increases
the computing load, in practice mostly 3-layer ANNs are preferred.
In this work, in the implementation stage of the ANN, Matlab 6.5 software is
used. In the program; three layers ANN model including one hidden layer with feed
forward and back-propagation algorithm has been trained by using Levenberg
Marquardt (LM) algorithm. The network used in this study has 12 neurons in the
hidden layer with the logarithmic sigmoid activation function which is non-linear
For inputs, along the peak load dataset, monthly temperature and population
growth are taken into account. In this study, the average monthly temperature
values are obtained from the regional meteorological record office. The monthly
population growth is calculated from the 1997 and 2000 national population
statistics using equation (16) which gives the population growth on monthly bases.
In this equation; P0 is the first and Pn is the second of the two consecutive
population statistics, n is the time interval between the statistics and r shows
population growth rate. The value of r is calculated as 5.04 x 10 -4. By using these
values the population is calculated on monthly bases. All the data which are used
for the training and testing of the ANN have been scaled to an interval of (0.25,
0.75).
In this paper, by using the peak data from TEDAS which was
recorded from 2000 2007 the least squares regression based method
and ANN, peak load demand for 2008 is validated. The results are
validated by using the real data of 2008.
PAPER 2
In this paper, the following models were used by the authors;
PROPOSED ELMAN RECURRENT NEURAL NETWORK
Fig. 2 portrays an Elman network, which is a subclass of recurrent neural
networks, has not only feed forward but also feedback connections and this
connection helps learning. Elman networks extend the static multilayer perception
(MLP) with context units; which are the PEs (processing elements) that remember
past activity. Context units are required when learning patterns over time i.e. when
the past values of the network influences the present processing. In the Elman
network, the output of the hidden PEs from the previous time step is copied to the
context units. In addition, the context units are locally recurrent i.e. they feedback
on themselves. The local recurrence decreases the values by a multiplicative
constant (time constant) as they are feedback. This constant determines the
depth i.e. how long a given value fed to the context unit will be remembered. One
can treat the context units as input unit, just as if they were obtained from external
source such as file.
AS
LOAD
PREDICTOR
FOR
architecture consists of five inputs. The number of PEs in the hidden layer is 10 and
that in the output layer is one.
connection, and a constant bias term, represented in the figure by the weight of a
connection with a fixed input equal to 1. The activation function must be a nondecreasing and differentiable function; the most common choices are the sigmoidal
(s-shaped) functions. An initially constructed ANN is like a newborn child. The
neuron weights are initialized with small random numbers when network is first
created. A process of learning or training is required to coach this unlearned
network by exposure to sample data. The most popular error correction algorithm is
the back propagation rule (BP). So, it will be used in this work.
PROPOSED TECHNIQUES
As mentioned before, long term load forecasts will be prepared using ANN
model and regression model. ANN model was used to get forecasts utilizing two
plans; two years plan and four years plan. Each plan consists of more than two
procedures we get different procedures by changing input variables and methods of
pre-processing. While regression model will be used for load forecasting of the next
year only Fig.2 shows a schematic diagram for proposed models.
REGRESSION MODEL
In this section we will use Linear Regression model to forecast peak load of
year 2006.Data used to prepare this model was historical data of years 1983-2005.
This model can be summarized as the following:
This equation will be applied for all data contained in data set, and load for
each year will be calculated. Figure 4 shows forecasted results from the proposed
model and the actual loads.
Here, the authors are concerned with long term load forecasting and
presents a comparison between two models applied to the Egyptian
unified network, these models are Artificial Neural Network (ANN) model
and regression model. Data pre-processing techniques have been applied
to improve forecasting accuracy of the model. Forecasting capability of
each approach is evaluated by calculating two separate statistical
evaluations of the Mean Absolute Percentage Error (MAPE) and the
Average Absolute Percentage Error (AAVE).
It can be seen from the results that the results attained with the
linear and exponential regression approach are very close to each other.
PAPER 2
EFFECT ON PERFORMANCE OF DAUBECHIES WAVELET WITH DIFFERENT
ORDERS AND LEVEL OF RESOLUTIONS
PAPER 3
1. For ANN model using GDP and population we get accurate forecasts with
MAPE 0.2104% with very simple two layer structure.
2. By changing the training set to use the new factor represented in
GDPPOP, we get more reduced error of 0.18%.
3. By applying this methodology to forecast both energy sales and peak load
for year 2007.We get forecasts with MAPE of 0.284.
4. Two (2) variable regression models were tested to forecast peak load of
year 2006. This model gives relatively good results, if compared to other
regression models proposed in literature. The forecasted load was 17837.8
MW, with percentage error 3.109%, the average error for 24 observations
equals 3.278 %.
SIGNIFICANT FINDINGS
PAPER 1
In this paper, upon comparison of the forecasting for 2008 results attained, it
was found out that with the use of longer input data, the forecasting error is
decreased. In this study, when the different load forecasting techniques are
compared for Kutahya, it is seen that the ANN approach has produced better results.
PAPER 2
CONCLUSION
This article review shows that among the methods available for
electric power load forecasting, Artificial Neural Network (ANN) based
methods gives the best result. It significantly reduces the forecasting
error.
This review also shows that combining two or more variables gives a
much better result in forecasting electrical load.
This also shows that when the available data has a longer time span,
the result of forecasting would be more accurate.
COMMENTS
Given that the area covered by the following research was very
large, still, the authors deliver a good result and suggests a much better
model in electric power load forecasting which is very helpful in another
study and if such method will be applied in a different area.
Data processing done by the researchers using MatLab software is
advantageous because of its built-in computing functions that can be
applied in load forecasting with ease and such reduces the risk of
computing errors.