Institutional Repository of Vilnius University
Trumpalaikis orų prognozavimas Lietuvos stotyse taikant grafų neuroninius tinklus /
Abstract
dc:descriptionThe aim of this thesis is to develop a probabilistic machine learning model based on observational data and adapted for point weather forecasting in a sparse and heterogeneous station network. To achieve this goal, historical observation data was prepared, a graph neural network with an attention masking mechanism was designed, and a latent neural process was integrated to evaluate forecast uncertainty. The developed algorithm is capable of directly processing incomplete measurements from 52 stations of the Lithuanian Hydrometeorological Service and does not require any artificial interpolation methods. The system generates forecasts up to 72 hours in advance for six different variables and simultaneously provides calculated confidence intervals. Testing results demonstrated that the proposed model achieves high structural accuracy and successfully outperforms baseline forecasting methods in the short term. This architecture is proposed as an effective alternative for generating short term local weather forecasts in areas characterized by limited measurement capabilities.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Timoščenka, Edvardas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308327178&prefLang=en_US