Study On Short-Term Passenger Flow Prediction Of Subway Considering Time Distribution Characteristics

As one of the most effective ways to relieve the pressure of ground transportation,urban rail transit has been attached great importance by local governments in recent years,and has set off a construction boom throughout the country.With the extensive coverage of rail transit network,its accessibility is improved and passenger flow continues to increase,and the change of passenger flow also presents an increasingly complex situation.As an important indicator to measure the operation status of the station,the passenger flow of rail transit can provide data support for the management department to formulate a reasonable operation plan,which is of great significance to give full play to the capacity of rail transit and ensure the safe and efficient travel of passengers.In this context,this paper takes the inflow passenger of a single station as the prediction object and sets 10 minutes as the prediction time granularity to study the shortterm passenger flow prediction method.The main contents are as follows:(1)Analyze the time distribution characteristics of inflow passenger of Hangzhou rail transit.Using K-means clustering algorithm to divide ‘week’ into working days and nonworking days.The frequency distribution and cumulative frequency of passenger flow are introduced,and 80% fractional of cumulative frequency is used as the critical value to divide non-peak and peak state.Using one-hot coding to discretize the features of working days,non-working days,non-peak and peak state,laying a foundation for the subsequent passenger flow prediction.(2)Controlling neural network parameters and input data,the prediction performance of LSTM,Bi-LSTM and mf-Bi-LSTM are compared.The experimental results show that mf-Bi-LSTM has the highest prediction accuracy,and this model is selected as the basic model of the combined model.(3)Based on variational modal decomposition(VMD),Sparrow search algorithm(SSA)and mf-Bi-LSTM,establish VMD-ISSA-mf-Bi-LSTM combined model.The VMD method is used to decompose the data into several intrinsic mode functions(IMF),and DTW distance and sample entropy were introduced as the indexes to identify the noise IMF.The noise IMF were eliminated and the remaining IMFs were the input data of the prediction model.The SSA algorithm is introduced as an optimization algorithm to find the optimal parameter combination of the prediction model,and its advantages and disadvantages are analyzed and improved.Finally,the improved SSA algorithm(ISSA)is used for parameter optimization.All IMFs were input respectively into the ISSA-mfBi-LSTM model for prediction,and sum up all prediction results to obtain the complete prediction result.(4)Carry out comparative tests.The same data set and prediction accuracy evaluation index were used to compare the prediction effect of VMD-ISSA-mf-Bi-LSTM model with VMD-mf-Bi-LSTM model,SSA-mf-Bi-LSTM model and ISSA-mf-Bi-LSTM model.Experimental results show that the prediction accuracy of the proposed combined model is significantly higher than other models,and the validity of the proposed model is verified.