Research On Inbound And Outbound Station’s Short-term Passenger Flow Prediction Method Of Urban Rail Transit

As a new,safe,efficient and green public transport tool,urban rail transit(URT)can effectively relieve traffic pressure,which could solve the imbalance between supply and demand in the transportation system in large and medium-sized cities to a large extent.The prediction results of short-term passenger flow provide an important foundation for decision-making in URT.Because accurate prediction results of short-term passenger flow of URT can ensure the operation safety and make the transport more efficient.The detailed research contents of this paper are as follows:(1)Taking into consideration the nonlinear,non-stationary,and other characteristics of the URT passenger flow time series,this paper used class empirical mode decomposition algorithm(EMD,EEMD,CEEMD and CEEMDAN)on passenger flow data.According to the noise reduction effect of passenger flow data,the CEEMDAN algorithm which has the best noise reduction decomposition effect is used to decompose the passenger flow time series data.The advantages and disadvantages of four kinds of EMD algorithms are analyzed,and the decomposed subsequences are used as the input variables of the neural network to prepare for the subsequent prediction.(2)In order to solve the problems of limited global search ability,slow convergence speed,and low convergence accuracy in the iterative process of particle swarm optimization(PSO)algorithm in searching for better results in neural network hyperparameters.In this paper,four key hyperparameters in long and short term memory(LSTM)neural networks are considered as optimization objects.An improved particle swarm optimization(IPSO)algorithm is proposed in this paper to dynamically solve the optimal value of a group of hyperparameters,enhance the local and global optimization capabilities of PSO algorithm,and further improve the prediction accuracy of LSTM.At the same time,the parameter optimization performance of IPSO algorithm is tested and analyzed by benchmark function,which verifies that IPSO algorithm achieves good optimization effect in the process of LSTM hyperparameters optimization.(3)Taking the inbound and outbound passenger flow data of Yangji Station of Guangzhou Metro in July 2019 as an example,the CEEMDAN-IPSO-LSTM combination model is constructed to accurately predict the short-term passenger flow of URT.The comprehensive analysis based on prediction error and the experimental results of the Taylor diagram show the following results.The data decomposition effect of EMD,EEMD,CEEMD and CEEMDAN is gradually improved.The IPSO algorithm has better parameter performance than PSO algorithm.The CEEMDAN-IPSO-LSTM model can more accurately solve the passenger flow prediction problem of URT in different periods than other models.