Pronóstico de temperaturas mínimas en todas las estaciones meteorológicas cubanas utilizando redes neuronales artificiales
Cargando...
Archivos
Fecha
Autores
Roque Rodríguez, Julio Cesar
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
En la actualidad los pronósticos de temperaturas extremas han llamado la
atención de especialistas e investigadores de diferentes campos, principalmente
en el área de las ciencias computacionales donde se han realizado un grupo
de investigaciones, demostrando que las técnicas de Aprendizaje Automatizado
(AA) ofrecen mejores resultados que los métodos convencionales.
A partir de información suministrada por el CMPVC y tratando de mejorar
los modelos utilizados en dicho centro, en el presente trabajo se utilizan las
“Redes Neuronales Artificiales” (RNAs) como una técnica eficaz para tal fin.
En la búsqueda de los modelos por estación o de modelos que se ajusten a la
mayor cantidad de las mismas se realizó una experimentación bastante extensa
con todas las estaciones teniendo en cuenta que cada una de estas debe ser
analizada para las dos temporadas del año (Invierno y Verano).
El potencial predictivo de estos modelos se analiza en función de: error medio
absoluto y de la desviación estándar, en búsqueda de una reducción de errores
durante la predicción de las temperaturas para una determinada estación
meteorológica. Los resultados finales se comparan con los del modelo MOS
(Model Output Stadistic) lo que muestra que los modelos propuestos ofrecen
predicciones más certeras para este tipo de datos en los que existe una relación
compleja entre ellos. La capacidad predictiva de los modelos de la aplicación
PronMLP resultó ser mejor en cuanto al% de casos positivos obtenidos, comparado
con el alcanzado por el modelo MOS en el período de tres años, desde
2/4/2012 hasta 20/4/2015.
At present extreme temperature forecasts have drawn attention of specialists and researchers from different fields, mainly in the area of computer science which have made a research group, showing that Automated Learning techniques (AA) offer better results than conventional methods. From information provided by the CMPVC and trying to improve the models used in the center, in the present study Artificial Neural Networks (RNAs) as an effective technique used for this purpose. In the search for models by station or models that fit as many of them quite extensive experimentation was carried out with all seasons considering that each of these must be analyzed for two seasons (winter and Summer). The predictive power of these models is analyzed according to: mean absolute deviation and standard error in seeking a reduction in errors when predicting temperatures for a given weather station. Final results are compared with MOS model (Model Output Stadistic) which shows that the proposed models provide more accurate for this type of data in which there is a complex relationship between them predictions. The predictive ability of the models of the application PronMLP proved to be better in terms of% of obtained positive cases compared to that achieved by the MOS model in the three-year period from 2/4/2012 to 20/04/201
At present extreme temperature forecasts have drawn attention of specialists and researchers from different fields, mainly in the area of computer science which have made a research group, showing that Automated Learning techniques (AA) offer better results than conventional methods. From information provided by the CMPVC and trying to improve the models used in the center, in the present study Artificial Neural Networks (RNAs) as an effective technique used for this purpose. In the search for models by station or models that fit as many of them quite extensive experimentation was carried out with all seasons considering that each of these must be analyzed for two seasons (winter and Summer). The predictive power of these models is analyzed according to: mean absolute deviation and standard error in seeking a reduction in errors when predicting temperatures for a given weather station. Final results are compared with MOS model (Model Output Stadistic) which shows that the proposed models provide more accurate for this type of data in which there is a complex relationship between them predictions. The predictive ability of the models of the application PronMLP proved to be better in terms of% of obtained positive cases compared to that achieved by the MOS model in the three-year period from 2/4/2012 to 20/04/201