To improve the accuracy and reliability of short-term prediction of highway
visibility level in key scenarios such as short duration and fast changing
speed, this paper proposes a short-term prediction method for highway visibility
level based on attention mechanism LSTM. Firstly, XGBoost and SHAP methods are
used to analyze the factors affecting highway visibility, determine the
importance ranking of different influencing factors, and select the factors that
have a greater impact on visibility as inputs for the visibility level
prediction model. Secondly, based on LSTM as the model foundation network and
innovative coupling attention mechanism, a visibility level prediction model
based on attention mechanism LSTM is constructed, which can dynamically update
the correlation between meteorological feature information at each historical
time point and the visibility level at the current prediction time, thereby
dividing the importance of information and flexibly capturing important
information in meteorological changes. The method proposed in the paper was
validated using data from 30 meteorological stations in China over the past 5
years, and compared with the commonly used BPNN, LSTM, and XGBoost methods for
visibility level prediction. The results showed that the overall prediction
accuracy of this method reached 88.8%, with an accuracy improvement of 16.7%
-30.8% compared to other methods, indicating good accuracy, and for each
visibility level, the prediction accuracy of this method is significantly better
than other methods, with good stability. Therefore, this method can effectively
predict the visibility level of highways and can be applied in practical
systems.