Analysis Of Short-term Passenger Flow Prediction Error Of Urban Rail Transit Based On Station Clustering And Time Granularity

In recent years,with the continuous expansion of subway scale,accurate short-term passenger flow prediction provides a powerful guarantee for reducing passenger waiting time and improving the efficiency of train operation organization.Scholars at home and abroad have done a lot of research on short-term passenger flow prediction methods for urban rail transit.However,there is still a lack of systematic research on the selection of objects for short-term passenger flow prediction and the selection of time granularity of prior data passenger flow.In view of this,this article will explore the differences in the accuracy of short-term passenger flow predictions for different types of subway stations based on the inbound passenger flow at different stations of the entire Xi’an urban rail transit network at different time periods.At the same time,a short-term passenger flow prediction method based on machine learning and empirical mode decomposition is proposed.First,based on the inbound passenger flow data obtained from the subway automatic fare collection system(AFC),8 types of passenger flow characteristics at different time periods were extracted as the site classification clustering factors,and the k-means method was used to divide the subway stations into 4 categories.In combination with the actual situation,the four types of stations are actually defined,which areⅠ(heavy residential station),Ⅱ(light residential station),Ⅲ(consumer tourism and passenger hub station),andⅣ(commercial work class).station).Secondly,the research scope of time is determined according to the passenger flow commitment rate within the effective period of subway operation.Within the selected time study range,the similarity measure of the Pearson coefficient is used to measure the number of pit stops at four different time granularities for the four types of stations.The optimal time granularity for passenger flow prediction is 15 minutes.In order to verify the validity of the time period selection for different types of stations,a more mature time series prediction model(ARIMA)for passenger flow prediction was used to predict 8 time periods.The prediction results show that the time periods with less prediction errors of the four types of stations are related to the data There is a direct relationship(the higher the Pearson coefficient for the time period with the smallest prediction error).Thirdly,the lookback volatility(LBV)is used to describe the fluctuation of passenger flow data,and a passenger flow prediction method based on empirical mode decomposition-support vector regression(HHT-SVR)is creatively proposed.Compared with the more mature ARIMA model,SVR model,and BPNN neural network prediction model,the prediction error obtained by the HHT-SVR prediction method has improved the prediction accuracy.Finally,using 87 metro stations in Xi’an as a sample,the data volatility LBV was fitted to the prediction error.The results show that the linear fitting results in a fitting effect R~2 of 0.812.It shows that there are obvious differences in the prediction errors of different types of stations.According to statistics,the average prediction errors of typesⅠtoⅣstations are:3.12%,4.61%,6.62%,and 3.89%.