Research On Forecast And Influencing Factors Of Vehicle Travel Time

With the progress of society and the development of the economy,car parc has been continuously improved,and the phenomenon of car congestion on urban roads has become increasingly serious.Intelligent transportation system is an important part of road traffic.At present,China's intelligent transportation is in the development stage,and the intelligent transportation system is constantly improving.The use of machine learning technology to analyze the vehicle travel time data can better understand the factors affecting the travel time of vehicles and predict the traffic conditions in a short time,so as to recommend the best route,improve the user's driving experience and save social costs.In this paper,the KDD CUP 2017 data set of the top events in the field of international data mining is used as empirical data,and machine learning technology is used to analyze the contribution of many factors affecting the travel time of vehicles.On this basis,a multi-model fusion algorithm is proposed,which can obtain higher accuracy of vehicle travel time prediction.Specific research contents including:(1)Data cleaning and exploratory analysis.We need deal with outliers and missing values in the data set.and the outliers are identified and treated as missing values for reasonable filling.According to the cleaning data,three kinds of features are constructed,including time feature,road feature and weather feature.(2)Analyze the influencing factors of vehicle travel time.Based on the constructed characteristic variables,the importance of these variables on the travel time of the vehicle is analyzed.The Random Forest(RF),General Gradient Boosting Regression(GBM)and Extreme Gradient Boosting Regression(Xgboost)are used to synthetically analyze the importance of influencing factors of vehicle travel time.The differences among the three models are compared according to the importance ranking,and the main factors affecting vehicle travel time are obtained according to the actual situation.(3)Prediction of vehicle travel time.The four algorithms of K-nearest neighbor,random forest,GBM and Xgboost are used to establish the vehicle travel time prediction models,then these models are combined by weighting method.The three evaluation indicators of MAE,RMSE and MAPE were used to contrastive analysis the prediction effect of the models.The results show that Xgboost is superior to other models in single models.The weighted fusion model is superior to single model in prediction,and the evaluation index is better,which is more suitable for short-term prediction of vehicle travel time.