Research On Recognition Algorithm Of GPS Track Loading And Unloading Points Of Freight Cars Based On DBSCAN Clustering And XGboost

The loading and unloading point has become a new high value of the track data after the stop point because of its axis function on the problems such as stroke segmentation,line feature characterization,service scope definition,resident management optimization,pain point shortage line discovery and so on.Different from the traditional track stop point,the loading and unloading point not only has the basic characteristics of position stay,time duration and so on,but also has the characteristics of the change of driving direction after the end of the order.The discovery of these change rules is of great value to the identification of the loading and unloading point.Compared with the traditional stop point recognition,it has the characteristics of scattered sparsity and speed dependence,and the variability of direction and speed increases the difficulty of loading and unloading point identification.This paper studies the identification of loading and unloading points in truck GPS trajectory based on DBSCAN clustering and XGBoost algorithm.The precise identification of loading and unloading points is realized by using the combination algorithm of DBSCAN and XGBoost twice,and the down-sampling algorithm process is also designed for the business scenarios such as dispatching order and assembling order,so that the filtering and election rules of data can be adjusted according to the business characteristics,and the error data which is difficult to be classified by the model can be eliminated effectively.the article also designed four control experiments in empirical research: one DBSCAN identification experiment,two DBSCAN identification plus downsampling identification experiments,one DBSCAN plus XGBoost identification experiment,two DBSCAN plus XGBoost and downsampling experiments.By comparing the experimental results obtained by different algorithms,the validity of the algorithm is verified,so as to select the optimal algorithm combination,and at the same time,the hyperparameters are adjusted by classifying the evaluation index results in the course of the experiment.The experimental results show that the classification effect of the combined algorithm model with two DBSCAN plus XGBoost and downsampling is moreeffective.Using this algorithm,the loading and unloading points in the GPS trajectory can be accurately identified,and the recognition effect in the test set is 87.81%recall,85.58% precision and 86.73% measurement.These experimental results show that the proposed algorithm can effectively solve the loading and unloading point problem of vehicle trajectory data and provide a new algorithm practice for solving similar problems.