Thesis DISEÑO, IMPLEMENTACIÓN Y VALIDACIÓN DE UN MODELO PREDICTIVO DE LA CONCENTRACIÓN DE OZONO TROPOSFÉRICO EN SANTIAGO BASADO EN MÉTODOS DE APRENDIZAJE AUTOMÁTICO
Date
2017
Authors
GONZÁLEZ LÓPEZ, IVÁN EDUARDO
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A pesar de que el Ozono troposférico se encuentra de forma natural en el aire respirable, enniveles más altos que los que se establecen en las normas nacionales vigentes, origina seriosproblemas para la salud humana y el entorno natural. La situación se suele agravar en lasgrandes zonas urbanas, donde debido a la actividad humana, hay una mayor concentración delos precursores de este contaminante. Por lo tanto, se hace necesario contar con herramientascapaces de modelar y predecir la concentración del Ozono troposférico, con el fin de proveerla información necesaria a las autoridades correspondientes y así puedan llevar a cabo deforma oportuna, las acciones preventivas pertinentes.Este trabajo propone un modelo de predicción para el nivel de Ozono troposférico en la ciudadde Santiago de Chile, basado en Redes Neuronales Recurrentes LSTM, utilizando comopredictores diversas variables de tipo químico y/o meteorológico. Los resultados muestranque este tipo de redes alcanzan un rendimiento competitivo respecto a los obtenidos por unaRegresión Lineal, Support Vector Machines (SVM), las Redes Neuronales Feed Forward ymodelos ARIMA para series de tiempo.
Although tropospheric ozone is found naturally in the breathable air, at higher levels than thoseestablished in the current national regulations, it causes serious problems for human healthand the natural environment. The situation is usually aggravated in large urban areas, wheredue to human activity, there is a greater concentration of the precursors of this pollutant.Therefore, it is necessary to have tools capable of modeling and predicting the concentrationof tropospheric ozone, in order to provide the necessary information to the correspondingauthorities so that they can carry out in a timely manner the relevant preventive actions.This work proposes a prediction model for the tropospheric ozone level in the city of Santiagode Chile, based on Recurrent Neural Networks LSTM, using as predictors various variablesof chemical and meteorological type. The results show that this type of networks achieve acompetitive performance with respect to those obtained by Linear Regression, SVM, FeedForward Neural Networks and ARIMA models for time series.
Although tropospheric ozone is found naturally in the breathable air, at higher levels than thoseestablished in the current national regulations, it causes serious problems for human healthand the natural environment. The situation is usually aggravated in large urban areas, wheredue to human activity, there is a greater concentration of the precursors of this pollutant.Therefore, it is necessary to have tools capable of modeling and predicting the concentrationof tropospheric ozone, in order to provide the necessary information to the correspondingauthorities so that they can carry out in a timely manner the relevant preventive actions.This work proposes a prediction model for the tropospheric ozone level in the city of Santiagode Chile, based on Recurrent Neural Networks LSTM, using as predictors various variablesof chemical and meteorological type. The results show that this type of networks achieve acompetitive performance with respect to those obtained by Linear Regression, SVM, FeedForward Neural Networks and ARIMA models for time series.
Description
Catalogado desde la version PDF de la tesis.
Keywords
APRENDIZAJE AUTOMATICO , LONG SHORT-TERM MEMORY , OZONO TROPOSFERICO , REDES NEURONALES RECURRENTES