Research On Short-term Prediction Method Of Bus Passenger Flow Based On Combination Model

With the increasing improvement of urban informatization level,there are more and more ways and means to obtain big data.How to use these data to analyze the bus passenger flow,deal with the changes of short-term passenger flow in the future,and make reasonable arrangements for bus dispatching is particularly important.The short-term prediction of bus passenger flow can provide passengers with accurate and real-time passenger flow information.Passengers can effectively avoid the peak passenger flow,improve the efficiency of passenger travel,and also provide an effective basis for alleviating urban traffic.In this paper,the obtained multi-source data of public transport are cleaned and fused,and the boarding stations of passengers are identified by hierarchical clustering method,so as to obtain the data of public transport passenger flow;Considering the weather conditions,air quality index,historical passenger flow,whether working days and whether peak hours,determine the input eigenvalue of the short-term prediction model of public transport passenger flow,and analyze the law of public transport passenger flow from three aspects:time,space and short-term distribution characteristics of public transport passenger flow;The first mock exam is linear selection and ARIMA model.The LSTM model and GRU model with strong nonlinear fitting ability are selected.Combined with the advantages of single model,two combined modes are adopted:series combination model and parallel combination model based on reciprocal error method.Four combined models of series ARIMA+LSTM combination model,series ARIMA+GRU combination model,parallel ARIMA+LSTM combination model and parallel ARIMA+GRU combination model are obtained.The prediction objects are bus line 1 passenger flow and railway station passenger flow.The time granularity is divided into 5 minutes and 15 minutes.The parameters of the model structure are determined by python.The error evaluation indexes are compared through root mean square error RMSE,average absolute error MAE and relative percentage error MAPE to illustrate the prediction accuracy of the model under different prediction objects and different time granularity.In the research process of short-term prediction method of bus passenger flow,the prediction accuracy of the model is as follows:parallel ARIMA+GRU combination model>parallel ARIMA+LSTM combination model>series ARIMA+GRU combination model>series ARIMA+LSTM combination model>GRU>LSTM>ARIMA;When the time granularity is 5 minutes,it can better predict the passenger flow of bus lines and stations.For the passenger flow of No.1 bus line,the best prediction time granularity is 5 minutes,and compared with ARIMA,the RMSE,MAE and MAPE of parallel ARIMA+GRU combination model based on error reciprocal method are reduced by 56.010%,56.524%and 49.041%respectively;For the passenger flow at the railway station,the best prediction time granularity is 5 minutes.Compared with ARIMA,the RMSE,MAE and MAPE of parallel ARIMA+GRU combination model based on error reciprocal method are reduced by 69.487%,70.479%and 74.670%respectively.