Asosiy kontentga oʻtish
proceeding

Modeling moisture hotspot formation in large warehouses using spatio-temporal graph neural networks

Anvarjon RashidovBukhara State Pedagogical Institute, Bukhara, UzbekistanMaksud MakhmudovBukhara State University, Bukhara, UzbekistanSamad NurillaevKhushvakht UmurovaBukhara State University, Bukhara, UzbekistanFeruza AbdullayevaBukhara State Pedagogical Institute, Bukhara, Uzbekistan
2026en

Ochiq kirish: yoʻq

ABI

Annotatsiya

We study the formation and early evolution of moisture hotspots in large warehouses through a computational framework that couples physics-based modeling with spatio-temporal graph neural networks. A reduced-order heat-mass transfer model provides priors on vapor diffusion, sorption, and airflow-driven advection. At the same time, the warehouse is represented as a graph whose nodes are sensor clusters and aeration zones and whose edges encode transport pathways. The ST-GNN integrates these priors via physics-informed regularization and conserves nonnegativity and mass balance. Uncertainty is quantified through probabilistic outputs calibrated to sensor noise and model discrepancy. Evaluations on multi-season operational data from high-capacity grain storage show that the proposed model anticipates hotspot onset with longer lead time and fewer false alarms than autoregressive baselines and purely spatial or purely temporal deep models. Ablation studies confirm the benefit of graph connectivity learned from airflow maps and of the physics constraints. The approach enables actionable early warnings and supports digital-twin what-if analyses for ventilation scheduling and risk-aware inventory management.

Hali tarjima qilinmagan

Identifikatorlar

Iqtiboslar va manbalar