Tesis doctorales: TESEO

Ayuda Salir

Ficha de Tesis

Imprimir

Ministerio de Universidades - Gobierno de España

Título: GENETIC ENSEMBLE (G-ENSEMBLE): AN EVOLUTIONARY COMPUTING TECHNIQUE FOR NUMERICAL WEATHER PREDICTION ENHANCEMENT
Autor: W. Y. Ihshaish, Hisham
Universidad: Universidad Autónoma de Barcelona
Departamento: ARQUITECTURA DE COMPUTADORES Y SISTEMAS OPERATIVOS
Fecha de Lectura: 12/09/2012
Programa de doctorado: Computación de Altas Presetaciones
Dirección:
- Senar Rosell, Miquel Àngel (DIRECTOR)
- Cortés Fité, Ana (DIRECTOR)
Tribunal:
- GARCÍA FERNÁNDEZ, INMACULADA (PRESIDENTE)
- PAZ GALLARDO, ABEL FRANCISCO (SECRETARIO)
- Alexandrow, Vassil (VOCAL)
Descriptores:
- COMPUTACION HIBRIDA
- MODELOS NUMERICOS DE LA ATMOSFERA
- INFORMATICA
Localización: BIBLIOTECA DE COMUNICACIÓN Y HEMEROTECA GENERAL UAB
Marcador: https://www.educacion.gob.es/teseo/mostrarRef.do?ref=992325Añadir a favoritos.
Resumen:
The main goal of the presented work is to tackle the problem of accuracy and waiting time in weather forecasting, which are normally conducted by computational applications known as Numerical Weather Prediction (NWP) models. These models have been strongly developed in the last decades and their performance constantly increases with the advances in computational power. However, in practice, many serious are still gaining considerable efforts by the scientific community in order to reduce what is widely known as "weather limited predictability". Mainly, the major two challenges are the willingness to get more reliable weather predictions, and to do it faster. As in many other areas of environmental modeling, NWP models, most simulation software works with well-founded and widely accepted models, the need for input parameter optimization to improve model output is a long- known and often-tackled problem. Particularly in such environments where correct and timely input parameters As in many other areas of environmental modeling, most simulation software works with well- founded and widely accepted models. Hence, the need for input parameter optimization to improve model output is a long- known and often-tackled problem. Particularly, in such environments where correct and timely input parameters cannot be provided. Efficient computational parameter estimation and optimization strategies are required to minimize the deviation between the predicted scenario and the real phenomenon behaviour. Based on the before mentioned, this thesis intends to: 1. Provide a sensitivity study of the effect of NWP model input parameters on prediction quality. 2. Propose a valid framework, which allows to search for the most 'optimal' values of model input parameters which, in our hypothesis, will provide better prediction quality. 3. Reduce the waiting time needed to get more reliable weather predictions. To accomplish the objectives of the presented proposal, a new weather prediction scheme is introduced. This new scheme implements an evolutionary computing algorithm, which focuses on the calibration of input parameters in NWP models The presented scheme is called Genetic Ensemble, which is composed of two-phases: calibration phase and prediction phase. Through the calibration phase, the presented approach applies Genetic Algorithm operators iteratively, in order to find 'best' values of NWP model input parameters, which consequently, will be used in the consequent prediction phase. Many strategies of the Genetic Ensemble have been developed, as such, it's extended to calibrate more than one level of input parameters, and also to evaluate their values using different strategies. On the other hand, the proposed scheme is paralleled using a Master/Worker programming paradigm, and is suitable to be executed in high performance computing (HPC) platforms, by which, execution time is intended to be reduced. The presented scheme has been evaluated by running weather prediction experiments over a well-known weather catastrophe; Hurricane Katrina 2005. Obtained results showed both significant improvement in weather prediction quality, and a considerable reduction in the over all execution time.

Imprimir

Modificar Consulta Nueva Consulta

Versión 4.4