Research On Passenger Flow Forecasting Method For Urban Rail Stations Based On EMD-KNN

The time series changes of passenger flow at urban rail transit stations show a certain regularity over a long period of time.However,if affected by factors such as large passenger flows during holidays or unexpected events,the changes in passenger flow at stations will be different than usual,and Passenger flow presents more non-linear and non-stationary characteristics.Precisely predict the size of passenger flow in the next period of time,which will help travelers grasp the information of passenger flow changes in real time and provide a reference for travelers’ travel planning;for rail transit operation managers,it is helpful to grasp real-time and accurate passenger flow changes.It allocates human resources reasonably,optimizes diversion facilities,and takes emergency measures.Empirical Mode Decomposition(EMD)can decompose complex nonlinear and non-stationary sequence signals into several relatively stable sequences,and K-nearest neighbor non-parametric regression(KNN)is widely used in time series pattern matching prediction.For this reason,this article mainly discusses the applicability of the combined prediction method(EMD-KNN)of empirical mode decomposition and K-nearest neighbor nonparametric regression in the time series prediction of inbound passenger flow at rail transit stations.First,this paper summarizes the research status of empirical mode decomposition method and K-nearest neighbor algorithm respectively,and introduces the principles of the two algorithms in detail.Then,taking the hourly inbound passenger flow time series of Guangjinan Road Station of Suzhou Rail Transit as the research object,the K-nearest neighbor non-parametric regression prediction method is used to predict the passenger flow at various times of several days.The algorithm steps mainly include the establishment of the database,the selection of the state vector,the determination of the neighboring K value,the weighted prediction and so on.Among them,the historical data is divided into two prediction scenarios of working days and non-working days,and the average absolute percentage error change rate is introduced to improve the efficiency of determining the neighbor K value.The results show that the K nearest neighbor algorithm has better prediction accuracy.Further,the empirical mode decomposition method and the K-nearest neighbor algorithm are combined for prediction.Under the premise of not distinguishing between working days and non-working days,the empirical mode decomposition of the passenger flow time series is carried out to obtain a number of eigenmode functions(IMF)and a residual sequence,and the components are grouped reasonably to obtain a high-frequency sequence,Low-frequency sequence,and trend item sequence.For each recombination sequence,K-nearest neighbor algorithm is used to predict,and the predicted value obtained from each recombination sequence is arithmetic superimposed to obtain the final predicted value.Forecast accuracy.Finally,analyze the changes in passenger flow at Guangjin South Road Station under the influence of the new coronavirus pneumonia(COVID-19)epidemic,and use the BP structure multi-breakpoint detection method to identify the approximate location of the structural breakpoint as January 23,2020.March 11,2020 and May 2,2020.The three structural breakpoints divide the time series into four sub-periods,and the KNN algorithm,the EMD-KNN combined algorithm and the ARIMA model are used to predict the passenger flow in the case of structural breakpoints.The results show that the change trend of the prediction results of the EMD-KNN combined algorithm and the single KNN algorithm is highly consistent with the true value,and the EMD-KNN combined algorithm has higher prediction accuracy and is more suitable for the non-linear and non-stationary hourly inbound passenger flow of rail transit Time series forecasting.