| To alleviate urban traffic congestion,current research is gradually shifting from ITS to ATS,and accurate prediction of bus travel time is a key link in the implementation process of ATS.However,the current accuracy of bus travel time prediction is not high enough,which seriously restricts the improvement of the operational efficiency of urban transportation systems.Therefore,in order to fully utilize the transportation resources of the road network,plan the transportation system more reasonably,and improve the level of dynamic path planning,this article conducts research on the prediction method of bus travel time based on bus operation data.Firstly,analyze the spatiotemporal variation characteristics of bus travel time,study the time variation characteristics of bus travel time on workdays and rest days on complete bus operation routes,and propose an improved dimensionless fluctuation quantification index to measure the stability of bus operation status.Secondly,based on fuzzy clustering to divide the bus operation status,the fuzzy C-means algorithm,K-means algorithm,OPTICS algorithm,and HAC algorithm are used to cluster the bus operation status.The number of clusters,SC,and DBI are used as evaluation indicators,and the fuzzy C-means algorithm with excellent clustering performance and practical needs is selected as the clustering method.Then,establish a bus travel time prediction model,combine sparrow search algorithm and long short memory network,construct a single station bus travel time prediction model,study the feature vectors,step processes,and parameter settings of the model,select model evaluation indicators and compare the models;On the basis of the single stop bus travel time prediction model,combined with the fuzzy C-means algorithm,a multi stop bus travel time prediction model is constructed,and the training process,evaluation index selection,and comparison model selection of the model are studied.Finally,to verify the accuracy and practicality of the model,the predictive performance of the single station bus travel time prediction model was verified from five aspects:different operating periods,different operating date properties,different weather conditions,different operating regions,and different model prediction time.The selected evaluation indicators were compared with the comparative model;Validate the predictive performance of the multi-station bus travel time prediction model from two aspects:long-distance multi-station and short-term multi-station.Compare and analyze the selected evaluation indicators with the single station bus travel time prediction model.Select the FCM algorithm to divide the operating state of buses into saturated flow state,comfortable flow state,and free flow state.Comparing the SSA-LSTM single station bus travel time prediction model with BP neural network,SVM,and LSTM,the results showed that the R~2 of SSA-LSTM was higher than that of BP neural network,SVM,and LSTM in different scenarios,and other indicators were also relatively good,proving that SSA-LSTM is more suitable for predicting the travel time of single station buses.Comparing the FCM-SSA-LSTM multi-stop bus travel time prediction model with the SSA-LSTM single stop travel time prediction model,it is proven that the R~2 of FCM-SSA-LSTM is 0.04 higher than that of SSA-LSTM over long distances,and the other errors of FCM-SSA-LSTM are generally smaller than those of SSA-LSTM model;At short distances,the R~2 of FCM-SSA-LSTM and SSA-LSTM are the same,and their prediction performance is equivalent.FCM-SSA-LSTM is more suitable for predicting bus travel time between multiple stations than SSA-LSTM. |
