Comparative Study on Statistical Prediction Methods of the Minimum Road Surface Temperature in Winter
BBased on the observation data of two traffic weather stations at Xinghe service area and the exit of the Wofoshan Tunnel along the G6 Highway from 2018 to 2020,the daily variation characteristics of the minimum road surface temperature in winter were analyzed,and the prediction models of winter minimum road temperature under simple feature scheme and complex feature scheme were established and tested by four statistical forecasting methods,including stepwise regression,regression tree,least squares support vector machine and deep neural networks.The results showed that the minimum road surface temperature was close to the air temperature at night in winter,and the road surface temperature increased rapidly in the morning,and reached the peak at about 13:00,which was earlier than the air temperature.The model test results showed that the complex feature scheme was better than the simple feature scheme in four evaluation indicators at these two traffic weather stations.According to the overall evaluation index,the model established by deep neural network method had the best prediction effect.The prediction accuracy and the average absolute error of the complex feature scheme at Xinghe service area were 74.72％and 2.53 ℃,respectively,while the prediction accuracy and the average absolute error at the exit of Wofoshan Tunnel were 96.38％and 1.14 ℃,respectively,which meant this model could be applied to forecasting road surface temperature.
万方数据
维普
