본문 바로가기
  • Home

A Physics-Informed Neural Network (PINN)-Based Model for Localized Fog Prediction

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(8), pp.21~28
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : June 20, 2025
  • Accepted : August 1, 2025
  • Published : August 29, 2025

Ji-Oh Jeong 1 In-Young Kim 1

1국방대학교

Accredited

ABSTRACT

Fog significantly impacts not only civilian sectors such as logistics and transportation but also the scale and strategy of military operations. However, its complex mechanisms and strong regional variability make accurate prediction challenging. This study proposes a localized fog prediction model that combines a Physics-Informed Neural Network (PINN) with regional meteorological data. Using weather observations from Taebaek, Gangwon Province (2015–2024), the model performs data-driven learning and incorporates three key physical relationships—Clausius-Clapeyron equation, relative humidity, and dew point—into the loss function to reduce error. Results show 84% accuracy and a 59% fog detection rate, surpassing the military’s empirical model and machine learning-based models such as XGB(49%) and LGB(50%). This study demonstrates that embedding physical principles into machine learning enhances the reliability of fog forecasting and supports the development of localized prediction models.

Citation status

* References for papers published after 2024 are currently being built.