Prediction Of Travel Time During Commuting Based On K-XGB Model

With the advancement of data collection technology and the development of data analysis methods,the use of traffic data in the past to predict future(short-term)traffic variables(Speed,Average travel time,Rate of lane occupancy)has become an important basis for the implementation of traffic management measures.The prediction of travel time in peak section can help travelers choose their driving route reasonably.It can also help the road regulators to identify the congested routes(links)and take the corresponding unblocking measures in advance to avoid the traffic congestion.Prediction of route travel time during commuting(early peak)is the focus of this thesis.This thesis first sorted out the papers in the field of short-term prediction.Next,identified the commuter-like vehicles in the peak period and defined them as T-car vehicles.Meanwhile,it explored the correlation between the proportion of T-car vehicles in the road section and the average travel time during the peak period of the day.Then,the thesis incorporated the proportion of this kind of vehicles into the prediction model to improve the accuracy of the prediction results.Finally,the thesis referred to the idea of KNN model fusion method and constructed a K-XGB travel time prediction model.This thesis collected the License Plate Recognition data of a prefecture-level city from July to November in 2013 for empirical analysis.The experiments verified that the K-XGB model has higher prediction accuracy than the traditional KNN model,the SVM model and the GBDT model.And as the prediction lead time increases,the prediction accuracy of the model has higher stability.The core work of this thesis is mainly divided into the following three parts: In the first part,the construction of the prediction model focused on the spatiotemporal characteristics of traffic data.Secondly,the article proposed a T-Car vehicle recognition algorithm and incorporated the proportion of such vehicles into the model prediction.The third point is to use the fusion idea of KNN to improve the prediction accuracy of the travel time prediction model based on XGBoost.The proposed model is expected to make full use of the characteristics of License Plate Recognition data.And the model focus on the trajectory information of the vehicle to improve the accuracy of the travel time prediction during commuting.