Forecasts of Road Surface Temperature in Winter Based on Random Forests Regression
Based on the data of three traffic meteorological stations set on Nanjing-Suqian-Xuzhou Expressway observed every ten minutes during 2015-2018, the random forests regression is used to forecast the road surface temperature in the next hour in winter and the feasibility and applicability of the models were analyzed. The results are as follows. The random forests regression method can be used to predict the road surface temperature of the expressway in winter, and the feature input scheme and the parameter debugging are different in different types of traffic meteorological stations. Compared with the simple features, the complex features can replenish and explain the environment and meteorological elements of the traffic meteorological stations better, and they have a higher degree of differentiation between the ordinary road traffic meteorological stations and the traffic meteorological stations near the bridge and water. Thus, the model has a good forecast effect on the general road traffic meteorological stations and the traffic meteorological stations near the water and bridges, but a little poor forecast effect on the traffic meteorological stations in the service areas. The reduction of the average error rate out of bag does not mean the improvement of the prediction accuracy. The random forest regression model simulated from the complex features can be used to predict the road surface temperature of different types of traffic weather stations in winter no matter in what weather conditions. The forecast effect is the best in rainy and snowy days, followed by in ouvercast days, but slightly worse in sunny days.