Aristotle University Of Thessaloniki (AUTH)
Πρόβλεψη απόκρισης καρστικών υδροφορέων με χρήση τεχνητών νευρωνικών δικτύων
Abstract
dc:descriptionThis doctoral thesis deals with simulation of the response of karstic springs by means of neural networks. the respective program has been constructed using rainfall and spring flow rate data from the area of the municipality of rouvas, located in the island of crete, greece. the thesis is divided into six chapters. αn appendix, including all field measurements and rainfall data concludes the thesis. in this thesis a feed-forward back propagation ann with 3 layers has been used. in feed-forward anns, the input signal propagates through the network in a forward direction, from layer to l to layer. the back-propagation learning algorithm was proposed by paul werbos in 1970. the procedure is supervised and it uses the so-called delta rule, namely it aims at minimizing the root mean square error (rmse) between expected and calculated results. it has already been mentioned that the ann used was based on fahlman’s quickprop algorithm. the sigmoid function has been used as activation function and rmse as the evaluation criterion regarding the structure of the ann, a three layer architecture has been selected. while the number of neurons of the input and output layers was determined by the selection of data and the desired output respectively, the number of the neurons of the hidden layer was decided by means of numerous tests. the fifth and main chapter of the doctoral thesis presents the application of artificial neural networks to the karstic aquifer of gergeri, which belongs to the municipality of rouvas. the data used are daily rainfall (since january 1, 2006) and flow rates of two springs in gergeri. the names of the springs are: “may vrisi” and “pera vrisi”. the measurements of the spring flow rates took place from april 16, 2007 to march 16, 2010 and they were conducted in the framework of this thesis. the aforementioned graphs show that the neural network performed very well. the curves of calculated flow rates followed closely those of the measured ones and exhibit similar trends. therefore we can conclude that the ann that we have constructed and trained, gave satisfactory results. another interesting case considered, is the possibility of estimating the flow rates of one spring, based on measurements of the other. so we investigated the possibility of estimating flow rates in pera vrisi using flow rate measurements of may vrisi. we observe from the graph of fig. 5.6, that the neural network gives very satisfactory results. the predicted flow rate values are close to measured ones, and that the respective curves exhibit similar trends.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2011
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Δαριβιανάκης, Νεκτάριος
- Darivianakis, Nektarios
Subjects
dc:subject × 12Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/27487