Short term and peak load forecasting using artificial neural network
Shivam A. Koli and Arya V. Joshi
Department of Electrical Engineering, walchand college of engineering, Vishrambag sangli, Maharashtra
416415, India
Abstract.
1.Introduction- because of nonlinear and non-stationary
relationships between the load data and
Short-term forecasts of electricity demand
weather variables. Therefore, an adaptable
are needed for control and scheduling of
technique is needed.
power systems, and hence they are very
An artificial neural network (ANN)
important for dispatchers. Short-term
approach is such an adaptable technique.
electric load demand is often considered as
The Artificial neural network uses a
a function of weather variables and human
supervised learning technique for training
social activities. Traditional econometric
neural networks. The goal of this type of
approaches establish functional
network is to create a model that correctly
relationships between weather variables
maps the input to the output using
and current load demand for the
historical data and various inputs so that
forecasting function, often assuming a
the model can then be used to produce the
linear relationship. However, as Park et al.
forecast. These load forecasts play a vital
(1991) indicate, the econometric approach
role in optimum unit commitment, start up
may not give sufficiently accurate results
and shut down of thermal plants, control of The output is a sum of the weighted
spinning reserve and buying and selling of nonlinear terms given by
power in interconnected systems.an
where
advantage of the Artificial neural network
is that it does not require any assumption
between load and weather variables in
advance. Moreover, the literature suggests
where Wᵢⱼ (i = l ,..., n ; j = l ,..., nᵢ) is the
that the accuracy of forecasts is better than
weight factor between the ith input node
traditional econometric forecasting.
and the ith hidden node; v ᵢ is the threshold
In this literature we propose a
value at the ith hidden node defined by
method of load forecasting based on a
the algorithm; and xⱼ is the jth input
single-layer neural network
element.
2.Artificial neural network
The learning of a network is a recursive
Artificial neural network is an information process which adapts the connection
paradigm that is inspired by biological strengths, i.e. the weights among the nodes
nervous systems. The system structure is and the threshold values at each iteration.
composed of a large number of highly The error function E is to update the
interconnected processing elements
(neurons) working together to solve
specific problems. Each unit operates only weights at each recursive minimum:
on its local data and on the inputs they
where K is the number of cases used to
receive via the connections. The proposed
train the network O(m) is the actual output
neural network consists of one input layer,
of the training case m, and Ô(m) is the
one hidden layer and one output layer. The
forecasted output of the network.
NN maps the input vector X to the output
vector Ô (Fig. 1). 3.selection of input variables
The most important work in building an
Artificial neural network load forecasting
model is the selection of input variables.
Ô Selection of inputs is based on judgement
of experience of the engineer and it is
carried out by trial and error since there is
no rule of thumb that can be followed to
determine the which inputs to use.
However, some statistical analysis can be
Figure 1:neural network
structure.
done to determine the variables with most associated with day of week D(n) where n
influence on power consumption. These changes from 1 to 7 representing days of
inputs are (a) hour and day, (b) weather week
associated inputs and (c) historical loads
3.1. hour indicator H(i)
3.3. weather indicating variables
Load changes throughout the day from one
hour to another. On a typical day in India
Figure 3: monthly power consumption and
load graph
Figure 2: daily load curve in India
Weather variables primarily include
peak load occurs generally between
temperature, wind velocity cloud
06:00PM to 08:00PM. And minimum load
coverage, rain etc. out of all the above
occurs at 03:00AM(Fig.2). We can
variables temperature seems to have
conclude that the magnitude differs
strongest correlation to power
between various hours by a large margin
consumption as displayed by Fig.4
there for an hour indicator H(i) (where i
changes between 1-24) is very helpful Although wind velocity and cloud
while calculating short term load 3.13.1 coverage only have small effect on load
hence no significant relation is seen.
3.2. Day indicator D(i)
therefore, only temperature as a weather
Load also changes throughout the week it
variable Is considered while calculating
is clear from the fig.3 that power
load forecasting
consumption in the weekend (day 6,7)
3.5. Normalized data
drops compared to power consumption in
the weekdays. To take this into account Normalization or scaling is not really a
while calculating load forecasting, variable functional requirement for the NNs to
learn, but it significantly helps as it
transposes the input variables into the data
range that the sigmoid activation functions
lie in (i.e. for logistic [0, 1] and tanh [-1,
1]. And it makes it so that all inputs are in weight update, called batch learning. This
comparable range study uses on-line training.in proposed
ANN backpropagation can be
Each input data is multiplied by an
implemented to effectively train the
amplitude and shifted by an offset. The
amplitude and offset are often referred to artificial neural network. Backpropagation
as normalization coefficients. The is an iterative algorithm, that helps to
normalization coefficients are computed minimize the cost function by determining
based on the minimum and maximum which weights and biases should be
values found across all of the data sets. adjusted. During every epoch, the model
The formulae used to compute these learns by adapting the weights and biases
coefficients are to minimize the loss by moving down
toward the gradient of the error. Thus, it
involves the two most popular
optimization algorithms, such as gradient
where Max(i) and Min(i) are the maximum descent or stochastic gradient descent.
and minimum values The input
5.limitations of ANN
components normalize the data using the
following formula: A major problem in modelling load
forecasting is that there are
usually many trends in the data.
4.Training of ANN For example, there may be a
seasonal trend reflecting the
Training is a process to find the input–
seasons of the year. As well as
output map by repeatedly analyzing the
there may be monthly and/or
training set as the network learns. The
weekly trends in the data as well.
parameters of the network obtain optimal
These trends make it difficult to
values during this process and the
correctly model load forecasting
performance checked with the test set. The
data.
ingredients for supervised learning are the
input, the desired response, the definition Artificial neural networks can
of error and a learning law. Training can be become too complex because of
implemented in two ways. Either we their architecture and the huge
present a pattern and adapt the weights datasets used to train them. They
(on-line training), or we present all the can also memorize the training
patterns in the input file (an epoch), data. These can lead to poor
accumulate the weight updates, then generalization of new data.
update the weights with the average
They can also be vulnerable to
adversarial examples or minor
changes in the input data. These
changes can cause the particular
artificial neural network to make
incorrect decisions and irrelevant
outcomes.
5. Conclusion
ANN based Short-term electric load and
energy forecast model has been developed
in this paper. ANN input variables have
also been chosen as a result of correlations
and contributions to decision analysis.
This developed model is extremely
valuable in planning fuel procurement
maintenance as well as ensuring that
supply meets demand. We have also
presented methods for training of such
ANN as well as the limitations that come
along with it.
7.Refrencees
