doktorska disertacija
Andrej Vidmar (Avtor), Mitja Brilly (Mentor), Matjaž Mikoš (Član komisije za zagovor), Mojca Šraj (Član komisije za zagovor), Ognjen Bonacci (Član komisije za zagovor)

Povzetek

Doktorska disertacija obravnava področje hidrologije in hidroloških modelov. Hidrološki pojavi se oblikujejo v hidrološkem sistemu, ki ga določa narava in so v osnovi stohastični. Pojavi so enkratni, neponovljivi in spreminjajoči se v prostoru in času. Simuliramo jih s pomočjo različnih hidroloških modelov, s katerimi poskušamo integrirati kompleksne procese, ki jih ponazarjajo zvezne funkcije. V preteklosti so bili razviti različni modeli, ki simulirajo različne hidrološke procese. Vendar so neskladja med simuliranimi in merjenimi vrednostmi še vedno velika in predstavljajo izziv mnogim raziskovalcem. Modeli vsebujejo veliko parametrov, ki jih je nemogoče neposredno izmeriti. Vrednosti večine teh parametrov so določene v postopku umerjanja, ki pogojuje učinkovitost takšnih modelov. Postopek umerjanja modela je obremenjen s pomanjkanjem podatkov in poenostavitvami, s katerimi rešujemo matematični model. Omenjene obremenitve povzročajo šume oziroma odstopanja med rezultati izračuna, ki nam jih poda model, in meritvami, ki smo jih pridobili v naravi. V postopku umerjanja modela poskušamo določiti parametre, ki jih vsebujejo posamezne funkcije, tako da so razlike med izračunom in meritvami čim manjše. Doktorska disertacija predstavlja uporabo izboljšanega postopka Gaussa-Levenberga-Marquardta (GLM) v kombinaciji s singularnim razcepom (SVD) in regularizacijo po metodi Tihonova. S takšnim postopkom izboljšamo umerjanja hidrološkega modela. Postopek se je preskušal na prosto dostopnem hidrološkem modelu z uporabo sintetičnih podatkov in na realnih modelih porečja Savinje v Sloveniji in reke Bosne v Bosni in Hercegovini. Na podlagi več kazalcev učinkovitosti umerjanja modela je bilo ugotovljeno, da postopek GLM v kombinaciji s SVD in Tihonovo regularizacijo zagotavlja učinkovito ujemanje zgodovine sintetičnega modela in skoraj popolno umerjanje parametrov. S pomočjo Tihonove regularizacije parametrov se je izdatno izboljšalo umerjanje modela in dosegla minimalna varianca napak. Poleg tega je bilo s primerjavo rezultatov predlaganega postopka z rezultati globalnih evolucijskih postopkov umerjanja ugotovljeno, da je samo umerjanje s kombiniranim postopkom GLM popolnoma simuliralo nizke pretoke. Ne nazadnje je bil šum pri rezultatih izračuna pri kombinirani metodi GLM z regularizacijo Tihonova praktično enak v postopku umerjanja in validaciji. To kaže, da je v rezultatih ostal le računski šum.

Ključne besede

Grajeno okolje;gradbeništvo;disertacije;delno distribuirani hidrološki model;učinkovitost umerjanja modela;GaussLevenberg-Marquardt postopek (GLM);singularni razcep (SVD);Tihonova regularizacija;vzporedno procesiranje;

Podatki

Jezik: Slovenski jezik
Leto izida:
Tipologija: 2.08 - Doktorska disertacija
Organizacija: UL FGG - Fakulteta za gradbeništvo in geodezijo
Založnik: [A. Vidmar]
UDK: 004.42:556(043)
COBISS: 52406531 Povezava se bo odprla v novem oknu
Št. ogledov: 532
Št. prenosov: 159
Ocena: 0 (0 glasov)
Metapodatki: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Ostali podatki

Sekundarni jezik: Angleški jezik
Sekundarni naslov: Calibration of semi-distributed hydrological models
Sekundarni povzetek: The doctoral dissertation deals with the field of hydrology and hydrological models. Hydrological phenomena are formed in a hydrological system determined by nature and are stochastic. Phenomena are unique, unrepeatable, and changing in space and time. We simulate these natural processes with the help of various hydrological models with which we try to integrate complex processes illustrated by continuous functions. In the past, different models have been developed to simulate different hydrological processes. However, discrepancies between simulated and measured values are still significant and pose a challenge to many researchers. The models contain many parameters that cannot be measured directly. The values of most of these parameters are determined in the calibration process, which is conditioned by the efficiency of such models. The process of model calibration is burdened by the lack of data and simplifications with which we solve the mathematical model. These all cause noise, or deviations between the results of the calculation given to us by the model and the measurements obtained in nature. In the process of calibrating the model, we try to determine the parameters contained in the individual functions so that the differences between the simulation and measurements are as small as possible. The doctoral dissertation presents the use of the enhanced Gauss-Levenberg-Marquardt (GLM) procedure in combination with singular value decomposition (SVD) and Tikhonov regularization. With such a procedure, we improve the calibrations of the hydrological model. The procedure was tested on a freely accessible hydrological model using synthetic data and on real models of the Savinja river basin in Slovenia and the Bosna river in Bosnia. Based on several model calibration efficiency indicators, it was found that the GLM process in combination with SVD and Tikhonov regularization ensures efficient matching of the history of the synthetic model and almost complete calibration of the parameters. With the help of Tikhonov parameter regularization, the calibration of the model was significantly improved, and a minimum error variance was achieved. Also, by comparing the results of the proposed process with the results of global evolutionary calibration processes, it was found that the calibration with the combined GLM process alone fully simulated low flows. Last but not least, the noise in the calculation results of the combined GLM method with Tikhonov regularization was practically the same either in the calibration or validation process, which shows that only the calculation noise remained in the results.
Sekundarne ključne besede: Built Environment;civil engineering;doctoral thesis;semi-distributed hydrological model;calibration model efficiency;GaussLevenberg-Marquardt procedure (GLM);singular value decomposition (SVD);Tikhonov regularization;parallel computing;
Vrsta dela (COBISS): Doktorsko delo/naloga
Komentar na gradivo: Univ. v Ljubljani, Fak. za gradbeništvo in geodezijo
Strani: XX, 151 str., 1 pril.
ID: 12585547