Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa
Thesis (MEng)--Stellenbosch University, 2025.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Stellenbosch : Stellenbosch University
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613906122309632 |
|---|---|
| access_status_str | Open Access |
| author | Van der Bank, Divan Kes |
| author2 | Bekker, B. (Bernard) |
| author_browse | Bekker, B. (Bernard) Van der Bank, Divan Kes |
| author_facet | Bekker, B. (Bernard) Van der Bank, Divan Kes |
| author_sort | Van der Bank, Divan Kes |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2025. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/134843 |
| institution | Stellenbosch University (South Africa) |
| last_indexed | 2026-06-10T12:43:35.067Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/134843 Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa Van der Bank, Divan Kes Bekker, B. (Bernard) Dalton, Amaris Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Wind -- Speed -- Forecasting -- South Africa Artificial intelligence -- Geophysical applications -- South Africa Weather forecasting -- South Africa Thesis (MEng)--Stellenbosch University, 2025. Van der Bank, D. K. 2025. Liberalizing Wind Speed Forecasting: A Limited-Area AI-based Approach for South Africa. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/672f96c4-62cc-400a-8ee2-273a0c910572 ENGLISH ABSTRACT: The increased penetration of variable renewable energy sources, such as wind power, poses challenges for transmission system operators and wind power producers in decisions regarding profitable market and grid balancing operations. Both parties would benefit from being able to generate accurate weather forecasts to mitigate the variable effects of wind speed and associated wind power changes. While Numerical Weather Prediction (NWP) is currently the cornerstone of weather forecasting, its high computational demands limit accessibility, making it primarily available to specialized weather forecasting companies and forcing wind power producers to subscribe to these companies for forecasts. Artificial Intelligence-based Weather Prediction (AIWP) has emerged as an alternative to NWP, providing forecasts of potentially similar accuracy while drastically lowering the computational costs required to generate forecasts. This advancement is driving the liberalization of the weather forecasting market from high development and operational computation requirements, enabling a broader range of users, including smaller energy producers and grid operators, to access and generate reliable, cost-effective forecasts. In particular, limited-area AIWP models (LAMs) are accelerating this transformation by offering high-resolution regional forecasts without needing large-scale supercomputing to develop the AIWP models. This thesis investigates three central hypotheses: (A) AI-based wind speed forecasting models can achieve forecasts with less than a 10% reduction in accuracy, as measured by standard validation metrics, while operating at significantly lower computational costs than traditional NWP models; and (B) limited-area AIWP models can provide higher temporal and spatial resolution forecasts at comparable accuracy and reduced development cost relative to global AIWP models. Once the LAM model is finalized and successfully proves hypothesis B, then the final hypothesis is that (C) the developed LAM can be operationalized into a real-time forecasting pipeline by (i) ingesting live weather data streams, and (ii) producing forecasts within timeframes suitable for operational decision-making. Focusing on wind speed forecasting in South Africa, this thesis validates that AIWP can be used as a computationally inexpensive alternative to NWP for medium-range wind speed forecasting. A custom evaluation platform is developed to standardize model comparisons specific to the South African region. To address the limitations in AIWP development, this thesis develops a custom LAM based on a graph neural network, significantly reducing the training and data processing demands of global AIWP models. The LAM produces forecasts that surpass the accuracy of Keisler’s model (which it is based on) with two key advantages: the LAM offers higher-resolution forecasts, both spatially and temporally, and has reduced computational overhead in producing the forecasts. Finally, the LAM is successfully integrated into an AI platform, demonstrating real-time operationalization and proving its suitability for decentralized, democratized access to accurate and timely wind speed forecasts. Through the process of testing the thesis hypotheses, it is ultimately shown that AIWP and LAMs can liberalize wind speed forecasting from high computational development and operational resource requirements, enabling more decentralized and liberalized access to accurate, timely wind speed predictions. AFRIKAANSE OPSOMMING: Die toenemende penetrasie van veranderlike hernubare energiebronne, soos windkrag, stel transmissiestelseloperateurs en windkragprodusente voor uitdagings rakende winsgewende mark- en netwerkbalanseringsbesluite. Albei partye sal baat vind by die vermoë om akkurate weer-voorspellings te genereer om die veranderlike effekte van windsnelheid en die gepaardgaande veranderinge in windkrag te verminder. Alhoewel Numeriese Weervoorspelling (NWP) tans die hoeksteen van weer-voorspelling is, beperk die hoë berekeningsvereistes daarvan die toeganklikheid, wat dit hoofsaaklik beskikbaar maak aan gespesialiseerde weervoorspellingsmaatskappye en die gemiddelde gebruiker dwing om op hierdie maatskappye te staatmaak. Kunsmatige Intelligensie-gebaseerde Weervoorspelling (AIWP) het na vore gekom as ’n alternatief vir NWP en bied voorspellings van moontlik soortgelyke akkuraatheid, terwyl dit die berekeningskoste wat nodig is om voorspellings te genereer drasties verminder. Hierdie vooruitgang dryf die liberalisering van die weervoorspellingsmark weg van hoë ontwikkelingsen operasionele berekeningsvereistes, wat ’n breër reeks gebruikers, insluitend kleiner energieprodusente en netwerkoperateurs, in staat stel om toegang te verkry tot betroubare, koste-effektiewe voorspellings. Beperkte-gebied AIWP-modelle (LAMs) versnel hierdie transformasie in die besonder deur hoë-resolusie streeksvoorspellings te bied sonder dat grootskaalse superrekenaars benodig word om die AIWP-modelle te ontwikkel. Hierdie tesis ondersoek drie sentrale hipoteses: (A) AI-gebaseerde windsnelheidvoorspellingsmodelle kan voorspellings lewer met minder as ’n 10% afname in akkuraatheid, gemeet aan standaard valideringsmetrieke, terwyl dit teen beduidend laer berekeningskoste as tradisionele NWP-modelle werk; en (B) beperkte-gebied AIWP-modelle kan voorspellings van hoër tydelike en ruimtelike resolusie lewer teen vergelykbare akkuraatheid en met laer ontwikkelingskoste relatief tot globale AIWP-modelle; en (C) die ontwikkelde LAM kan geoperasionaliseer word binne ’n intydse voorspellingspyplyn deur (i) regstreekse weerdata-strome in te neem, (ii) voorspellings te lewer binne tydraamwerke wat geskik is vir operasionele besluitneming, en (iii) akkuraatheid te handhaaf binne die drempels wat in (A) en (B) vasgestel is. Met ’n fokus op windsnelheidvoorspelling in Suid-Afrika, bevestig hierdie tesis dat AIWP as ’n berekeningkundig goedkoop alternatief vir NWP gebruik kan word vir mediumtermyn windsnelheidvoorspelling. ’n Pasgemaakte evalueringsplatform is ontwikkel om modelvergelykings te standaardiseer wat spesifiek is aan die Suid-Afrikaanse streek. Om die beperkings in AIWP-ontwikkeling aan te spreek, ontwikkel hierdie tesis ’n pasgemaakte LAM gebaseer op ’n graaf-neurale netwerk, wat die opleidings- en dataprosesseringsvereistes van globale AIWP-modelle aansienlik verminder. Die LAM lewer voorspellings wat die akkuraatheid van Keisler se model (waarop dit gebaseer is) oortref met twee sleutelvoordele: die LAM bied voorspellings van hoër resolusie, beide ruimtelik en tydelik, en het laer berekeningslas in die lewering van die voorspellings. Laastens is die LAM suksesvol geïntegreer in ’n AI-platform, wat intydse operasionalisering demonstreer en sy geskiktheid bewys vir gedesentraliseerde, gedemokratiseerde toegang tot akkurate en tydige windsnelheidvoorspellings. Deur die toetsing van die tesis-hipoteses word dit uiteindelik getoon dat AIWP en LAMs windsnelheidvoorspelling kan bevry van hoë ontwikkelings- en operasionele hulpbronvereistes, wat meer gedesentraliseerde en geliberaliseerde toegang tot akkurate en tydige windsnelheidvoorspellings moontlik maak. Masters 2026-01-12T10:20:18Z 2026-01-12T10:20:18Z 2025-12 Thesis https://scholar.sun.ac.za/handle/10019.1/134843 Stellenbosch University xiii, 103 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Wind -- Speed -- Forecasting -- South Africa Artificial intelligence -- Geophysical applications -- South Africa Weather forecasting -- South Africa Van der Bank, Divan Kes Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa |
| title | Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa |
| title_full | Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa |
| title_fullStr | Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa |
| title_full_unstemmed | Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa |
| title_short | Liberalizing wind speed forecasting: a limited-area AI-based approach for South Africa |
| title_sort | liberalizing wind speed forecasting a limited area ai based approach for south africa |
| topic | Wind -- Speed -- Forecasting -- South Africa Artificial intelligence -- Geophysical applications -- South Africa Weather forecasting -- South Africa |
| url | https://scholar.sun.ac.za/handle/10019.1/134843 |
| work_keys_str_mv | AT vanderbankdivankes liberalizingwindspeedforecastingalimitedareaaibasedapproachforsouthafrica |