Results - Details
Search command : Author="Κοσιώρης"
And Author="Γεώργιος"
Current Record: 3 of 10
|
Title |
Χρήση τεχνητών νευρωνικών δικτύων και στατιστικών τεχνικών για τη βελτίωση της πρόγνωσης μετεωρολογικών παραμέτρων |
Author
|
Θεολογίτου, Παρασκευή
|
Thesis advisor
|
Κατσαούνης, Θεόδωρος
|
Reviewer
|
Κοσιώρης, Γεώργιος
Τσιρώνης, Γεώργιος
|
Abstract |
Accurate weather forecasting is an integral part of everyday life and concerns both the
personal life and the industrial development of societies. The use of computational
models such as Artificial Neural Networks (ANN), which reduce errors and make
predictions valid, constitute the appropriate tools for successful forecasting. The aim of
this thesis is the setting up of an ANN to predict wind velocities, based on observational
and numerical data, for the city of Heraklion, Crete during severe weather events. The
comparison of different types of training algorithms of a Feed Forward ANN, such as the
Back Propagation Algorithm, the Random Optimization Algorithm as well as a hybrid
model consisting of the combination of the two previous algorithms, was set as a first
priority. Simultaneously, the effectiveness of the structural elements that make up the
ANN was examined. The analysis was performed by processing data from averaged daily
temperatures. Subsequently, using the most efficient ANN configuration, an attempt was
made for direct prediction of wind velocity in Heraklion, for days when severe weather
events occurred. The numerical data were from WRF model simulations, whereas
observational data were obtained from a surface weather station situated in the Physics
Department Building of the University of Crete. Moreover, with the aid of statistical
analysis, an attempt was made to predict a correction factor for the wind velocity which,
by applying it to the numerical values, will improve the prediction. For this reason, an
algorithm was developed to improve the lag in time that is observed between numeric and
observations for the examined events. Finally, the results were evaluated, and conclusions
were drawn on the ability of the ANN to efficiently predict the values of the wind
velocities. Also, suggestions are made for future directions of research in order to
optimize the performance of the ANN.
|
Language |
Greek |
Issue date |
2020-07-24 |
Collection
|
School/Department--School of Sciences and Engineering--Department of Physics--Graduate theses
|
|
Type of Work--Graduate theses
|
Permanent Link |
https://elocus.lib.uoc.gr//dlib/5/6/9/metadata-dlib-1594628863-319288-9154.tkl
|
Views |
247 |