Abstract:

To improve the prediction accuracy of winter road surface temperature, an hourly prediction model of road surface temperature of the expressway in winter based on multi-dimensional LSTM neural network was proposed. The hourly road surface temperature as the model output, the sliding window was used to construct a feature input matrix by considering the cumulative effect of multi-dimensional meteorological factors on pavement temperature and the periodicity of road surface temperature. The hourly road surface temperature prediction model based on LSTM was constructed to efficiently approximate the road surface temperature with complex nonlinearity and uncertainty by deep learning and was validated with examples of Ning-Su-Xu expressway in Jiangsu province and Ma-Zhao expressway in Yunnan province. The results show that compared with the random forest(RF) model and the BP neural network model, the hourly road surface temperature prediction accuracy of the proposed LSTM model is significantly improved, in which the mean absolute error (MAE), mean square error (MSE) and root mean square error (RMSE) of the LSTM model in Ning-Shu-Xu expressway and Ma-Zhao expressway are 0.303, 0.295, 0.543 and 0.581, 0.694, 0.833 respectively, and theabsolute error between the predicted value and the observed values is between [0, 1) ℃, accounting for 93% and 89%. The LSTM model accurately captures the periodicity and uncertainty of road surface temperature, with good model robustness, when the predicted values on rainy and sunny days are basically consistent with the measured values.

Key words: expressway, road surface temperature, long short-term memory(LSTM), hourly forecasting, meteorological factors