Surrogate-based optimization methods for coastal aquifer management

Christelis, Vasileios; Χριστέλης, Βασίλειος

dc.contributor.author	Christelis, Vasileios	en
dc.contributor.author	Χριστέλης, Βασίλειος	el
dc.date.accessioned	2021-03-17T07:04:41Z
dc.date.available	2021-03-17T07:04:41Z
dc.date.issued	2021-03-17
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/53074
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.20772
dc.rights	Default License
dc.subject	Παράκτιοι υδροφορείς	el
dc.subject	Seawater intrusion	en
dc.subject	Βελτιστοποίηση αντλήσεων	el
dc.subject	Pumping optimization	en
dc.subject	Μετα-μοντέλα	el
dc.subject	Surrogate models	en
dc.subject	Βελτιστοποίηση πολλαπλής πιστότητας	el
dc.subject	Surrogate-based optimization	en
dc.subject	εξελικτικοί αλγόριθμοι	el
dc.subject	Multi-fidelity optimization	en
dc.title	Surrogate-based optimization methods for coastal aquifer management	en
dc.title.alternative	Μέθοδοι βελτιστοποίησης βασισμένες σε μετα-μοντέλα στην διαχείριση παράκτιων υδροφορέων	el
dc.contributor.department	ΤΟΜΕΑΣ ΕΡΓΩΝ ΥΠΟΔΟΜΗΣ ΚΑΙ ΑΓΡΟΤΙΚΗΣ ΑΝΑΠΤΥΞΗΣ, ΕΡΓΑΣΤΗΡΙΟ ΕΓΓΕΙΟΒΕΛΤΙΩΤΙΚΩΝ ΕΡΓΩΝ ΚΑΙ ΔΙΑΧΕΙΡΙΣΗΣ ΥΔΑΤΙΚΩΝ ΠΟΡΩΝ	el
heal.type	doctoralThesis
heal.classification	Γεωεπιστήμες & Επιστήμες Περιβάλλοντος	el
heal.classification	Επιστήμες Ηλεκτρονικών Υπολογιστών και Πληροφορικής	el
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2021-02-05
heal.abstract	In pumping optimization of coastal aquifers, often the objective is to mitigate the phenomenon of saltwater intrusion while satisfying the demands for freshwater extraction. This management task is typically formulated as a nonlinear constrained optimization problem where the individual pumping rates are the decision variables. Evolutionary algorithms are considered highly competent to find a near-optimal solution to this difficult optimization problem, at the expense of thousands of objective function evaluations with the physics-based seawater intrusion model. This approach leads to an impractical computational cost when it is implemented using the high-fidelity but computationally expensive variable density flow and solute transport models. To that end, the present thesis focused on the use of surrogate models or metamodels as a realistic approach to the computationally expensive problems of pumping optimization of coastal aquifers. The aim was to develop new surrogate-based optimization (SBO) methods, that can realistically be applied in real-world coastal aquifer management problems. Radial basis functions and Kriging surrogate models were selected to develop SBO frameworks. Emphasis was given on the development of online SBO methods where the accuracy of the surrogate models was further enhanced in an iterative fashion (infill strategy). The following SBO schemes were developed: (a) A surrogate-assisted evolutionary framework using a pure exploitation infill strategy. (b) A surrogate-assisted evolutionary framework based on a multiple surrogate approach (either selecting the best surrogate model or forming an optimal weighted ensemble of surrogate models) and a pure exploitation infill strategy. (c) An adaptive-recursive optimization framework which balances exploration with exploitation using the metamodels. (d) Multi-fidelity optimization based on seawater intrusion models of different fidelities and co-Kriging models. In overall, results showed that the proposed SBO methods drastically reduced the computational cost of the corresponding optimization based on the variable-density flow and solute transport model. For pumping optimization problems of moderate dimensionality (i.e., 10 decision variables), all methods approached the region of the global optimum while those balancing exploration with exploitation provided near-optimal solutions within just 100 evaluations with the variable density flow and solute transport model. For pumping optimization problems of a larger dimensionality (i.e., 20 decision variables), the performance of the pure exploitation infill strategies, was negatively affected. However, the optimization schemes which balance exploration and exploitation had a consistent performance and located near-optimal solutions within 300 evaluations with the variable density flow and solute transport model. The new SBO method that balances exploration with exploitation, and it was developed in this thesis, performed better or comparable to other published SBO algorithms that have been successfully applied in the water resources optimization literature. The proposed multi-fidelity optimization framework delivered good local solutions by using as few as 21 simulations with the high-fidelity variable density flow and solute transport model and 200 simulations with a low-fidelity sharp interface model for a pumping optimization problem of 10 decision variables. The multi-fidelity optimization method outperformed conventional SBO for such a limited number of simulations with the high-fidelity seawater intrusion model.	en
heal.sponsor	Η διατριβή υποστηρίχτηκε με υποτροφία από τον Ειδικό Λογαριασμό Κονδυλίων Έρευνας (Ε.Λ.Κ.Ε.) του ΕΜΠ	el
heal.advisorName	Μαντόγλου, Αριστοτέλης	el
heal.committeeMemberName	Καρατζάς, Γεώργιος	el
heal.committeeMemberName	Νάνου-Γιάνναρου, Αικατερίνη	el
heal.committeeMemberName	Κατσιφαράκης, Κωνσταντίνος	el
heal.committeeMemberName	Μπαλτάς, Ευάγγελος	el
heal.committeeMemberName	Ναλμπάντης, Ιωάννης	el
heal.committeeMemberName	Τσιχριντζής, Βασίλειος	el
heal.committeeMemberName	Μαντόγλου, Αριστοτέλης	el
heal.academicPublisher	Σχολή Αγρονόμων και Τοπογράφων Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	166 p.	en
heal.fullTextAvailability	false