Expressway Visibility Prediction in Complex Terrain Based on LSTM-Attention Mechanism
Year: 2026
Type: conference-paper
Abstract: Low-visibility weather such as localized fog poses a severe threat to expressway traffic safety, especially in complex terrain areas. Traditional meteorological models fail to capture the abrupt non-linear spatiotemporal variations of visibility, and often overlook the modulation of static topographic features. To address these gaps, this paper proposes a novel visibility prediction model based on Long Short-Term Memory (LSTM) network, integrated with temporal attention mechanism and static geographic constraints. Based on high-resolution meteorological grid data and DEM topographic data of the G60 Expressway (Jiangxi section), the model integrates elevation, slope, and core physical drivers of visibility evolution, adapting to both mountainous and plain terrains. Experiments show that the proposed model significantly outperforms BPNN, SVR and vanilla LSTM benchmarks. It mitigates long-sequence information dilution, achieves a Threat Score (TS) of 0.78 for severe low-visibility events (<500m), and accurately captures critical precursors (e.g., nocturnal radiation cooling) before visibility abrupt drops.
Authors: Jichang Long, Yu Zhou, Xiaohui Huang, Xun Tian
Language: English
Subfield: Atmospheric Science
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7213531963