National Technical University of Athens (NTUA)
Βελτιστοποίηση αντλήσεων παράκτιων υδροφορέων με περιβαλλοντικά και οικονομικά κριτήρια
Abstract
dc:descriptionThe main goal of the dissertation is to develop efficient methods for pumping optimization of coastal aquifers. This is important in integrated management of water recourses in coastal areas. The research focuses on two general coastal aquifer optimization cases. In the first, the objective is to maximize pumping subject to environmental constraints for protecting the aquifer from seawater intrusion. In the second case, it is assumed that the sustainable aquifer yield is not sufficient to satisfy demand and the objectives are simultaneous minimization of economic and environmental cost, while satisfaction of demand is set as a constraint. The solutions are based on a simulation - optimzation methodology. An optimization algorithm proposes systematically possible solutions which are evaluated from physically based models that simulate the coastal aquifer regime. While stochastic optimization algorithms are effective, their combination with computationally demanding variable density models, leads to a formidable optimization task. For this reason, surrogate models are incorporated into the simulation - optimization methodology in place of the physical models of the system. Surrogate models based on Neural Networks are very popular and were investigated here. In large problems with many decision variables, training of neural networks is computationally extremely demanding, and the trained model had no good generalization ability. In order to alleviate this problem, we proposed using Modular Neural Networks (MNN) as surrogates. In order to reduce the unknown parameters of the MNNs, an adaptive geometrical technique is developed that allows eliminating unnecessary inputs from each module. This reduces significantly the complexity of the MNN and the training time. Further, original optimization algorithms are developed, where MNNs are combined with stochastic optimization algorithms. In this methodology, the MNNs are employed to propose possible optimum solutions, and only the most promising ones are evaluated using the numerical model. This has the advantage that the errors of the surrogate models do not have a negative impact on the final solution of optimization. The algorithms are applied to hypothetical examples and to real aquifers in two Greek islands. The proposed methods required only 5-7% of the optimization time of the standard simulation-optimization approaches, while the results are similar or even better than the previous methods.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kourakos, Georgios
- Κουράκος, Γεώργιος
Subjects
dc:subject × 20- Υφαλμύρυνση
- Μονοκριτηριακή βελτιστοποίηση
- Πολυκριτηριακή βελτιστοποίηση
- Μοντέλα αντικαταστάτες
- Τεχνητά νευρωνικά δίκτυα
- Γενετικοί αλγόριθμοι
- Μοντέλα μεταβλητής πυκνότητας
- Seawater intrusion
- Single objective optimization
- Multi objective optimization
- Surrogate models
- Meta-models
- Artificial neural networks
- Modular neural networks
- Genetic algorithms
- Variable density models
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Πολιτικού Μηχανικού
- Engineering and Technology
- Civil Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/18238