Szczegóły publikacji
Opis bibliograficzny
Time-slices based training of physics-informed neural networks for 3D non-stationary thermal inversion simulations / Maciej SIKORA, Maciej PASZYŃSKI // W: Computational Science – ICCS 2026 workshops : 26th International Conference : Hamburg, Germany, June 29–July 1, 2026 : proceedings , Pt. 1 / eds. Maciej Paszyński, Amanda S. Barnard, Yongjie Jessica Zhang. — Cham : Springer, cop. 2026. — ( Lecture Notes in Computer Science ; ISSN 0302-9743 ; LNCS 16786 ). — ISBN: 978-3-032-29911-6; e-ISBN: 978-3-032-29912-3. — S. 53–66. — Bibliogr., Abstr. — Publikacja dostępna online od: 2026-06-26
Autorzy (2)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 168890 |
|---|---|
| Data dodania do BaDAP | 2026-08-31 |
| DOI | 10.1007/978-3-032-29912-3_5 |
| Rok publikacji | 2026 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Wydawca | Springer |
| Konferencja | International Conference on Computational Science 2026 |
| Czasopismo/seria | Lecture Notes in Computer Science |
Abstract
During the day, the temperature in urban areas rises with height, leading to a negative vertical temperature gradient that causes the warm air near the ground to rise. While daytime convection facilitates mixing, nocturnal radiative cooling creates thermal inversions that trap emissions near the surface. This paper models this transition using time-dependent, three-dimensional advection-diffusion equations solved via Physics-Informed Neural Networks (PINNs). We propose a sequential training strategy over a four-dimensional space-time domain, partitioning the temporal axis into discrete segments to ensure convergence. In this framework, the terminal state of each segment provides the initial conditions for the subsequent slice, maintaining physical continuity. Our findings demonstrate that this PINN-based approach effectively captures the accumulation of ground-level pollutants during inversion events, achieving high numerical accuracy with significantly lower computational overhead compared to traditional grid-based Eulerian solvers.
