Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 7

Matrix of Literature

Long-short term Memory


Principal component
Author/s (Year) Energy Forecasting Recurrent Neural Normalization Activation Functions
analysis
Networks
Energy forecasts are
getting more and more
complicated as more
(Gonzalez volatile renewable sources
Ordiano emerge. Different tools        
et al., 2018) and services are offered
and tend to increase as
time goes by as it will be
significant in the future.
In addition, a data-driven
prediction do not need to
  be explicitly programmed        
as the data itself has the
information implied in it
According to the source,
the prediction of energy,
particularly solar and
wind, is greatly affected
by the weather. But the
preciseness of weather
(Aurangzeb forecasts is heavily
       
et al., 2019) dependent on some
factors like temperature,
humidity, and time lag. In
the same manner, an
evaluation index like
MAPE is considered
when forecasting.
Energy forecasting today
is complex and its
interpretation cannot be
(Jacob done one human alone.
       
et al., 2010) To solve this issue, high-
resolution data are
preferred for better
analysis and evaluation
Energy forecasting today
is complex and its
interpretation cannot be
done one human alone. To
solve this issue, high-
resolution data are
  preferred for better        
analysis and evaluation.
In addition, a data driven
prediction do not need to
be explicitly programmed
as the data itseld has the
information implied in it.
With regards to the
application of
normalization to
long short term
memory RNN,
(Jacob
et al., 2010)
suggested the last
three inputs to be
         
normalized. This
resulted to low
MAPE,
although the change
was small and
improvements
mostly came from
changing
algorithms.
Forecasting plays an
important role in the
operations and planning
Modern ways are
of the power system, that
suggested in recent years
is why improving the
to attain an accurate
accuracy of forecasting
measurement forecast.
(Liu et al., would be vital. It is not
Thus, computers are used      
2018) linear and non-stationary.
to run intelligence-based
In the same manner, an
simulations like long short
evaluation index like
term memory recurrent
MAPE or mean absolute
neural networks.
percentage error is
considered when
forecasting
LSTM can go over
(Hochreiter & minimal time lags thus it
Schmidhuber,   can solve complex      
1997) problems like time series
prediction.

two LSTM was used to


encode and decode
the series of words
and translate English into
(Sutskever, are converted into a
French. Their study was
Vinyals, & Le,       new vector with a
able to extract the
2014) constant
sequence of words and
dimensionality
turn it into a vector with
fixed dimensions.

Although the LSTM


seems promising, the
vanishing gradient
problem arises because of
(Stanley &
  the repeated cycle of the      
Taraza, 2001)
input data experiences.
With this problem, the
output of an LSTM model
deteriorates.
the researchers utilize an
Amazon Web Service
known as Software
(Aibin, 2019)        
Defined Elastic Optical
Networks that includes
LSTM.

On that note, SELU,


also known as "scale
exponential linear
unit", is a kind of
activation function
the normalization that belongs to the
of input data is self normalizing
Sola, J.
    essential before neural networks. This  
Sevilla, J.
training an artificial kind of activation
neural network function has an ability
to achieve a zero
mean and unit
variance even though
it already underwent
many network layers.

introduced Self
Normalizing Neural
(Costa et al.,
    Networks to    
2013)
combat high
variance
Principal component
analysis or PCA is a
technique that can be
(Rich, 2012).        
used to reduce the
number of features or
dimensions of data.

PCA uses rows and


columns, just like a
matrix, for the
(Vanderwarker, variables and
Amber M.,         components. In the
Marcoux, 2014) same sense, it also
provides numeric
outputs known as
component scores

PCA is almost used


by all scientific
disciplines since it
(Abdi &
can extract vital
Williams,        
insights from data and
2010),
simplify the
complexity of the
data.
Activation functions
tell what would be the
outcome of a neuron
(Pomerat,
in an artificial neural
Segev, & Datta,        
network and it
2019)
establishes the non-
linearity in a neural
network.

Based on the work of


about activation
functions, ReLU or
rectified linear unit is
the best choice as
         
compared to the two
common activation
functions: the sigmoid
and hyperbolic
tangent.

discussed the
advantage of vReLU
or V-shaped ReLU
compared to
(Hu, 2018)        
LReLU, ELU, and
SELU using MNIST
and not-MNIST
datasets
Energy forecasting will
never stop. It will just be It must be done to Comparison between
more complex and Forecasting is a time lessen the time other activation
(Tolentino , complicated as the series, thus, LSTM can be required for a functions shall be reduce dimentionality
2020) technology develops applied to it. network to function done of data

You might also like