Research On Traffic Flow Forecasting Method Based On Gradient Boosting Decision Tree

With the rapid development of intelligent traffic system(ITS),a variety of traffic information monitoring network brings together a lot of traffic data.It is very important to carry out data mining and explore the potential traffic behavior.Real-time and accurate traffic flow forecasting is one of the main contents of traffic data mining,which can not only predict the traffic congestion situation,improve the road capacity,but also provide scientific decision-making theory basis for the traffic management.Traditional traffic flow forecasting model mostly uses the static historical data with simple attribute and small size as training sample,which cannot describe the complex characteristics of traffic data and meet the demands of long-term traffic flow forecast.In order to solve the above problems,this paper combines the idea of integrated learning,and uses the gradient boosting decision tree to forecast the traffic flow based on data preprocessing,multidimensional temporal series analysis and temporal feature construction.Firstly,the original traffic data are classified,counted and filled,the time window is designed based on the multidimensional temporal series analysis method.The similarity measure is used to measure the similarity degree.According to the similarity ranking of the time window and the forecast window,combined with the rules of week,select the most similar k time window,that is,k learners.Then,the temporal difference sequence,temporal trend sequence and temporal deviation sequence are constructed for the n-dimensional original data of the most similar time window and the prediction window through the temporal feature construction algorithm to enhance the feature diversity of the sample as the input of the gradient boosting decision tree.Finally,the gradient boosting decision tree is used to forecast the traffic flow and output k predictions.According to the combination strategy,the simple average of k predictions is taken as the final traffic flow forecast value.In this paper,experiments are carried out with real data set.Experimental results show that,using gradient boosting decision tree after data preprocessing,multidimensional temporal series analysis and temporal feature construction can forecast the medium and long term traffic flow for large scale traffic data and obtain higher prediction accuracy.