Study On The Forecast Of Road-surface Temperature In Winter On North Part Of Nanjing-Suqian-Xuzhou Expressway

Low temperature of road surface in winter is the precondition of road icing and it has great importance in monitoring,forecasting and research for road safety.Accurate prediction of road surface temperature in winter can not only improve the technology and service level of meteorological department,but also improve the efficiency of road maintenance.What's more,it can also remind drivers of driving safety and ensure the safety of people's lives and properties.This essay analyzed the characteristics and established the framework of the next hour forecast product of road surface temperature in winter in the north section of Nanjing-Suqian-Xuzhou expressway based on the real-time observation data of AWMS in Jiangsu,geographic information and thermal data with machine learning(Random Forest Regression)and geostatistical interpolation(Tension Spline Interpolation).It was expected to lay a foundation for the warning and prediction of road surface temperature and road icing in winter of the whole expressway network in Jiangsu,provided the technical support for the use of machine learning to forecast the road surface temperature,and provided reference for the realization of refinement and intelligent traffic meteorological service in China.The major results could be summarized as follow:(1)The annual average road surface temperature in winter of the north section of Nanjing-Suqian-Xuzhou expressway was higher than 0?.The temperature at night was lower than that in the daytime.The temperature of the traffic meteorological station in the north was lower than that in the South.The temperature of the traffic meteorological stations near the service area was higher than that of the ordinary traffic meteorological stations,and more higher than those near the bridge and water.(2)Random Forest Regression can be used to forecast the road surface temperature in winter of the north section of Nanjing-Suqian-Xuzhou expressway but the feature input and the parameter debugging were different among different types of traffic meteorological stations.The complex features supplemented the explanation of the environment and meteorology of the traffic meteorological stations better so that the model simulated by that feature input showed remarkable prediction effect.The results of the ordinary traffic meteorological stations and the traffic meteorological stations near the bridge and water forecasted by the model were more precise than those of the traffic meteorological stations near the service area.(3)As to the forecast of the spatial road surface temperature of the north section of Nanjing-Suqian-Xuzhou expressway,the Tension Spline Interpolation combined with Random Forest Regression simulated by the complex feature input showed remarkable prediction effect.The parameter debugging schemes were that the parameter could be chosen if the average error out of bag equaled 1.0? for the traffic meteorological stations near the service area and if the average error out of bag equaled 1.5? for the ordinary traffic meteorological stations and the traffic meteorological stations near the bridge and water.The forecast of the south section of expressway was slightly better than that of the north section.(4)The forecast product of road surface temperature in winter of the north section of Nanjing-Suqian-Xuzhou expressway was consist of the single station forecast model and the spatial interpolation forecast model.Random Forest Regression was the core of the single station forecast model whose input features were visibility,temperature,relative humidity,precipitation,2-min average wind direction,2-min average wind speed,road surface temperature,10 cm Subgrade Temperature,D-value between temperature and 10 cm subgrade temperature,1-hour cumulative precipitation,1-hour temperature change,1-hour relative humidity change and 1-hour 10 cm subgrade temperature change and outputs were the next hour road surface temperature of each traffic meteorological station.Tension Spline Interpolation was the core of the spatial interpolation forecast model whose input was the result of the single station forecast model and output was the continuous road surface temperature in the whole road section.The road surface temperature of a certain point could be inquired with its latitude and longitude.