Page 138 - IJEEE-2022-Vol18-ISSUE-1

134 |
In this paper, weather-sensitive loads are predicted by
five components. Lighting and Domestic (Domestic
component refers to electrical appliances used throughout
the year, such as television, radio, dishwasher...etc.) These
components are not affected by weather conditions. The
cooling component is influenced by high temperatures.
While Air Heating and Water heating components are
influenced by the temperature drop. An artificial neural
network was used due to its high efficiency in electrical load
prediction.
A. Artificial Neural Network
Artificial Neural Networks are a good option for
predicting loads, due to their ability to find a complex non-
linear relationship between load and factors affecting it. An
artificial neural network (ANN) is designed to simulate the
way the human brain processes data. They are quite different
from statistical methods of analysis. An artificial neural
network builds its knowledge by discovering patterns and
relationships in data and by learning or training, not by
programming. The ANN technique is used to find the
relationship between multiple input variables and output
variables when it is difficult to find the relationship between
them mathematically. It highlights the importance of ANNs
in classification, pattern recognition, prediction, modeling,
and automated control. Artificial neural networks do not
require knowledge of the data source but require large
training sets [25]. Network structure consists of nodes
(neurons) connected by links and usually organized into
many layers. Each node in the layer receives and processes
the weighted inputs from a previous layer and transmits its
output to nodes in the next layer through links. Each link is
assigned a weight, which is a numerical estimate of the
conduction force. The weighted aggregation of node inputs
is transformed into outputs according to the transfer function
(usually a sigmoidal function). Most ANNs have three or
more layers as shown in Fig. 1.
Fig. 1: Deep neural network with 3 hidden layers.
The first layer, the input layer, is used to deliver data to
the network (contains nodes equal to the number of input
variables). The last layer, the output layer, is used to output
the output variables (contains nodes with the same number
Al-Nasiri, Al-bayaty & Al-Hafidh
of output variables). and one or more intermediate layers,
which are used to act as a set of isotropic (hidden layers) [26].
B. Artificial neural network Modeling
The artificial neural network created using MATLAB.
After defining the training and testing data set, the neural
network is created and the advantages of this network are
determined, including the type of network, number of hidden
layers, number of their neurons, transfer function, training
function, and the method of evaluating the performance of
the network. A feed-forward backpropagation network
created, with thirteen neurons in the input layer and ten
hidden layers, has ten neurons in each hidden layer, five
neurons in output layers, refers to the forecast five-
component load, as show in Fig. 2.
Fig. 2: Feedforward neural network
Mean Absolute Percentage Error and Mean Squared
Error is used to evaluate network performance:
MAPE = ?in=1|Yi-YYi 'i| (1)
n
MSE = ?ni=1 (Yi-Y'i)2 (2)
n
where: Y: the actual load, Y': the forecast load, n: number
of samples, i: sequence of the day.
C. Case Study
In this research, Mosul in northern Iraq load data for the
city of were used to evaluate the performance of forecasting
models. Figure 3. shows the data of peak daily load and
weather data for a year. Readings started from April 1 to
March 31 2010.