Page 70 - Contributed Paper Session (CPS) - Volume 7

CPS2031 Javier Linkolk L. et al.
Organization (WHO) standards and Chilean national air quality standards
(NAQS) [8].
This pollution episodes are associated with weather conditions [9], mainly
in the cold season, and there is little understanding of how the variation in
particle matter differs between cities and how this is affected by the
meteorological conditions [10].
On the other hand, the interest in Machine Learning has exploded in the
last decade. Fundamentally, Machine Learning is the use of algorithms to
extract information from raw data and represent it through some kind of
mathematical model.
Under this context, the research aims to clusterize scenarios and
estimation of higher PM2.5 pollution in the "La Florida" area, categorized as
the zone with the highest pollution index in the Metropolitan Capital. For this,
machine learning techniques will be used, such as self-organized maps (SOM)
and long-term memory networks (LSTM).

2. Methodology
Self-Organizing Maps
The Self Organizing Maps (SOM) model was introduced by T. Kohonen
[11]. The model preserves the topology mapping from the high-dimensional
input space onto a low-dimensional display.
The Map M consists of an ordered set of prototypes wk E W c R , k=1...M,
d
with a neighbourhood relation between these units forming a grid, where k
indexes the location of the prototype in the grid. The most common used lattices
are the linear, the rectangular and the hexagonal array of cells. In this work we
will consider a rectangular grid where K(wk)=(i, j) E N is the vectorial location
2
of the unit wk in the grid, where i and j stand for the row and column of the
prototype in the rectangular array [12].
When the data vector x E Rd is presented to the model M, it is projected
to a neuron position of the low dimensional grid by searching the best
matching unit (bmu), i.e., the prototype that is closest to the input.
The learning process of this model consists in moving the reference vectors
towards the current input by adjusting the location of the prototype in the
input space. The winning unit and its neighbours adapt to represent the input.

Long Short-Term Memory (LSTM)
The LSTM are a type of recurrent neural network with memory over a
period of time by adding to "memory cell". As indicated in [13], LSTM RNNs
addresses the problem of the misclassification in RNNs incorporating
activation functions in their state dynamics. This "memory cell" is controlled
mainly by "the entrance door", "the door of forgetfulness" and "the exit door".

57 | I S I W S C 2 0 1 9