Abstract:

Using the records of winter expressway accidents occurring in Liaoning province and the corresponding multi-factor weather observational data from 2014 to 2016, the spatial distribution characteristics of expressway accidents influenced by meteorological conditions were analyzed.Two-step clustering method was used first in data mining analysis processing to determine the number of winter weather types in Liaoning province, and then the cluster analysis of the meteorological data was performed using K-means method.The random forest method was used to construct an expressway traffic accident classification model for different weather types, and finally, the characteristic importance of meteorological factors in the model was analyzed.The results show that the number of expressway accidents influenced by meteorological conditions is the highest in the southern Liaoning province, followed by the western area, and numbers in the eastern and northern area are relatively low.The winter weather in Liaoning province can be divided into four types.According to the data structure of meteorological factors, the weather characteristics are as follows:precipitation occurring on the day, on the previous day, cold and dry, and warm and humid.The accident rate of weather types with obvious precipitation characteristics exceeds 70%, and the one with cooling and warming weather types is about 20%.The random forest method has high classification accuracy for the two weather types, i.e., precipitation occurring on the day and cold and dry types.The generalization ability of the model is better.The characteristic importance parameters of meteorological factors for the four weather types are significantly different.The surface temperature factor has a greater significance, occupying the first place in the high-incidence weather types, and the third place in the humid warm weather type.

Key words: Expressway, High-risk weather, Data mining, Meteorological factors