Zonificación de una red hidrometeorológica mediante técnicas de reducción dimensional y algoritmos de clasificación automática
Cargando...
Archivos
Fecha
2021
Autores
Director
Título de la revista
ISSN de la revista
Título del volumen
Publicador
Páginas
Resumen
El presente proyecto presenta una propuesta metodológica para el análisis de rutas de viaje mediante una primera aproximación de la red de estaciones hidrológicas y meteorológicas pertenecientes al Instituto Costarricense de Electricidad (ICE). Su proceso de análisis consiste en la aplicación de técnicas de reducción de dimensionalidad, métodos de clasificación automática y, por último, algoritmos de aproximación de la ruta más corta con el objetivo de describir la red mediante distancias en carretera y tiempos de viaje. Como insumos para la realización de esta investigación, se utiliza la información geolocalizada de las estaciones hidrometeorológicas, datos referentes a las carreteras que componen la Red Vial Nacional de Costa Rica e información de duración en tiempo entre estaciones. Utilizando software de Sistema de Información Geográfica, se computa el algoritmo de Dijkstra para obtener una matriz de disimilaridades basada en distancias. Por otro lado, mediante los servicios de Google Maps API, se genera la matriz de disimilaridades necesaria para analizar el tiempo. Se establecen distintos escenarios para la descripción de la red, en los cuales se toman combinaciones de técnicas cómo: Escalamiento Multidimensional (MDS) clásico, MDS no métrico y t-SNE, con métodos de agrupamiento: k-Medias, k-Medoides, DBSCAN y OPTICS. Se concluye que se encuentran resultados que logran disminuir las distancias y tiempos de viaje por medio de t-SNE y k-Medias, también con la aplicación directa de k-Medoides sobre las estaciones. Finalmente, se plantea una técnica para eliminación de ruido en la aplicación de OPTICS en la que se obtienen agrupaciones que superan los resultados obtenidos mediante las técnicas tradicionales tanto en distancia como tiempo. Esta técnica queda sujeta a mayor experimentación para verificar su efectividad con diferentes bases de datos.
Inglés: This project presents a methodological proposal for the analysis of travel routes through a preliminary approximation based on the network of hydrological and meteorological stations belonging to the Costa Rican Institute of Electricity (ICE, from its Spanish acronym). Its analysis process consists of the application of dimensionality reduction techniques, automatic classification methods and, finally, shortest path approximation algorithms with the objective of describing the network by road distances and travel times. As inputs for this research, geolocated information from hydrometeorological stations, data regarding the roads that compose the Costa Rican National Road Network and time duration information between stations are utilized. Using Geographic Information System software, the Dijkstra algorithm is computed to obtain a dissimilarity matrix based on distances. On the other hand, using Google Maps API services, the dissimilarity matrix needed to analyze time is generated. Different scenarios are established for the description of the network, in which combinations of techniques are taken, such as: Classical MDS, non-metric MDS and t-SNE, with clustering methods: k-Means, k-Medoids, DBSCAN and OPTICS. Where, it is concluded that the results that manage to reduce the distances and travel times are obtained by the use of t-SNE in combination with k-means algorithm, also with the direct application of k-Medoids over the stations. Finally, a technique for noise elimination in the application of OPTICS is proposed in which we obtain clusters that exceed the results obtained by traditional techniques in both distance and time. This technique is subject to further experimentation to verify its effectiveness with different databases.
Inglés: This project presents a methodological proposal for the analysis of travel routes through a preliminary approximation based on the network of hydrological and meteorological stations belonging to the Costa Rican Institute of Electricity (ICE, from its Spanish acronym). Its analysis process consists of the application of dimensionality reduction techniques, automatic classification methods and, finally, shortest path approximation algorithms with the objective of describing the network by road distances and travel times. As inputs for this research, geolocated information from hydrometeorological stations, data regarding the roads that compose the Costa Rican National Road Network and time duration information between stations are utilized. Using Geographic Information System software, the Dijkstra algorithm is computed to obtain a dissimilarity matrix based on distances. On the other hand, using Google Maps API services, the dissimilarity matrix needed to analyze time is generated. Different scenarios are established for the description of the network, in which combinations of techniques are taken, such as: Classical MDS, non-metric MDS and t-SNE, with clustering methods: k-Means, k-Medoids, DBSCAN and OPTICS. Where, it is concluded that the results that manage to reduce the distances and travel times are obtained by the use of t-SNE in combination with k-means algorithm, also with the direct application of k-Medoids over the stations. Finally, a technique for noise elimination in the application of OPTICS is proposed in which we obtain clusters that exceed the results obtained by traditional techniques in both distance and time. This technique is subject to further experimentation to verify its effectiveness with different databases.
Descripción
Tesis (licenciatura en ingeniería civil)--Universidad de Costa Rica. Facultad de Ingeniería. Escuela de Ingeniería Civil, 2021
Palabras clave
ANALISIS DE RUTA (ESTADISTICA), ANALISIS MULTIVARIANTE - TECNICAS, CLASIFICACION AUTOMATICA - METODOS, HIDROMETEOROLOGIA - COSTA RICA, REDUCCION DE DIMENSIONES (ESTADISTICA) - TECNICAS