Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/03844E0D-F9D5-4C10-86C0-E5D736783EF8 | ||
| Year | 2021 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Georgios Plessias, "Assessment of spatial and temporal distribution of rainfall in Crete using neural networks", Diploma Work, School of Chemical and Environmental Engineering, Technical University of Crete, Chania, Greece, 2021 https://doi.org/10.26233/heallink.tuc.91151 | ||
| Appears in Collections |
The objective of the current thesis is using and training neural networks, in order to simulate rainfall in the region of Crete. Crete is located at the southernmost tip of the Aegean Sea and covers an area of 8 336 km², with hydrological observation stations covering its entire area. Neural networks are used to simulate rainfall data collected from hydrological stations. Initially, data processing was necessary, as it required the creation of both an input table and a target vector in the neural network. The input table consisted of the coordinates of the hydrological stations, the wind speed and the date. The target vector contains the actual rainfall values. It is worth noting that the data came from 46 hydrological stations and relate to the period from 1/2/2006 to 30/6/2021. With the completion of the data processing, the training of the neural networks with the Neural Fitting tool (nftool) and the Neural Network tool (nntool) began. Two different training algorithms were also used Bayesian-Regularization and Levenberg-Marquardt. The training of the neural networks was then carried out, differentiating with each iteration the parameters in terms of the tools used, the hidden nodes and the training algorithms. During the process of training artificial neural networks, the effort is focused on finding the model and its parameters that offer the best results. Model selection criteria include mean square error, correlation coefficient and Nash – Sutcliffe model efficiency coefficient (NSE). The training of artificial neural networks included all available hydrological stations. In conclusion, after the completion of the training of artificial neural networks, the best accuracy was achieved by using the Bayesian Regularization training algorithm and by using the Neural Network tool (nntool) to refer to the total number of stations and years of observation.
Copyright © DIAS 2013
