20.500.12556/RUL-175892 Napovedovanje pretokov slovenskih rek s strojnim učenjem magistrsko delo Forecasting the streamflows of Slovenian rivers using machine learning Naraščajoča pogostost in intenzivnost vremenskih ekstremov zaradi podnebnih sprememb povečuje potrebo po zanesljivem napovedovanju pretokov slovenskih rek, zlasti zaradi nevarnosti hudourniških poplav. Namen magistrskega dela je bil raziskati možnosti uporabe metod strojnega učenja za kratkoročno, 18-urno napoved pretokov na hidroloških postajah po Sloveniji. V analizo so bile vključene meritve pretokov in padavin Agencije Republike Slovenije za okolje(ARSO), ter vremenske napovedi modela ALADIN. Razvit je bil model globokega učenja TOK, ki poleg zgodovine pretokov in lokalnih meritev upošteva prostorske značilnosti padavin širšega zaledja. Rezultati na 119 hidroloških postajah na 77 rekah so pokazali, da TOK presega linearno regresijo in ARSO-ov obstoječi hidrološki prognostični sistem (HPS) tako v standardnih hidroloških metrikah, kot tudi v zgodnjem opozarjanju na hudourniške poplave. S SHAP vrednostmi so pojasnjeni glavni dejavniki, ki vplivajo na napovedi modela TOK. Delo prinaša izboljšan pristop k zgodnjemu opozarjanju na poplavne dogodke in prispeva k razvoju naprednih hidroloških napovednih sistemov v Sloveniji. The increasing frequency and intensity of weather extremes due to climate change highlight the need for reliable forecasting of river discharges in Slovenia, particularly because of the risk of flash floods. The aim of this thesis was to investigate the applicability of machine learning methods for short-term, 18-hour discharge forecasts at hydrological stations across Slovenia. The analysis included discharge and precipitation measurements from the Slovenian Environment Agency (ARSO) and weather forecasts from the ALADIN model. A deep learning model, TOK, was developed, which, in addition to discharge history and local measurements, also incorporates the spatial characteristics of precipitation over a wider catchment area. Results from 119 hydrological stations on 77 rivers showed that TOK outperforms linear regression and ARSO’s existing Hydrological Forecasting System (HPS), both in standard hydrological metrics and in early flash flood warning. Using the SHAP values, the main factors influencing the predictions of the TOK model were explained. The work provides an improved approach to early flood warning, and contributes to the development of advanced hydrological forecasting systems in Slovenia. globoko učenje napovedovanje pretokov napovedovanje poplav SHAP deep learning river discharge forecasting flood prediction SHAP true false false Slovenski jezik Angleški jezik Magistrsko delo/naloga 2025-11-13 08:15:16 2025-11-13 08:15:21 2025-12-16 03:47:38 0000-00-00 00:00:00 2025 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.8 156639 256692227 20940.pdf 20940.pdf 1 C8137F6F2CCB35184FA05A1800AD05D2 ed2836971ac65396958b95747d053b713b76e956c251b893eb54c26865b0d551 2fa56fc5-c060-11f0-9328-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=221829 Fakulteta za matematiko in fiziko 0 0 0