Analysis Of Driving Behavior Of Truck Driver And Risk Of Overspeed Assessment Based On Big Data

In China,road transport carries more than 30 billion tons of freight volume every year.It is one of the most important ways of transportation in China.At present,the freight transportation management platform plays the role of the bridge of drivers and goods.The freight driver groups are generally contain 1 to 3 people,which is accessed to the management platform makes it easily to analysis the driver’s performance and ensure their safety.With the improvement of supervision on vehicle drivers,it is necessary to the transportation management platform to record and analyze vehicle transportation safety and driving behavior.Transportation platform works hard to use the historical transport data to analysis and ensure transportation safety.This paper was based on the data of a freight transportation platform in Guangdong Province,it established an evaluation model of freight transportation driving behavior.Through the analysis of data characteristics,the risk of speeding in different road sections was obtained which made it possible to forecast the potential risk of speeding in different road sections.Firstly,this paper cleaned the historical data which are recorded by the freight vehicle on-board terminal and stored in the database.It contains location,speed,mileage and alarm information.And the Lagrange interpolation method was used to replace or fill the outliner and missing data in the database.After that,11 fields related to driving behavior were extracted,and three feature information was further derived from the above 11 fields information.Then we used Box-Cox transformation to improve the Characteristics of skew distribution of the features.Secondly,the characteristic data were clustered twice by K-menas++ algorithm,.After that the drivers of transport vehicles were clustered into five categories.Combined with the existing driving behavior evaluation index and enterprise demand,the K-means++ evaluation basis of transport driving behavior was proposed as the label of driver rating.Then,this paper introduced a variety of ensemble learning methods.The Catboost ensemble learning algorithm was used to model and analyze the driving behavior of freight vehicle driver for the first time.The results showed that the Catboost algorithm has the best performance compared to the GBDT and DNN algorithm,with the accuracy and precision of 0.88 and recall of 0.86.Then,through the importance analysis of Catboost model,it was concluded that speeding behavior is one of the key factors affecting driving safety.Last,HMM algorithm was used for trajectory matching of vehicle location data,made each data record is mapped to the corresponding road section.And then we used LSTM neural network and RNN neural network to model and predict the daily trend of over-speed rate.The results showed that LSTM has a good prediction effect on over-speed rate of the road,because LSTM overcomes the disadvantage of RNN neural network which gradient vanishes easily for long time series prediction.