Research On Prediction Of Subway Inbound And Outbound Passenger Flow Considering Outliers And Spatial Features

In recent years,the concept of "smart transportation" has been constantly mentioned,and public transportation is developing towards a convenient,efficient and intelligent direction.As the top one contributor to alleviating urban traffic congestion,subway has naturally become an important part of the smart transportation network,and the call for intelligent construction of it is also increasing.Although the construction of underground railway network is developing vigorously in China,more improvement and optimization in terms of railway line layout planning,station setting,operation management and control still needs to be done.It is necessary to learn from developed countries.The cornerstone of all this is the ability to accurately predict passenger flow.As the world’s second most densely populated country,Singapore should have acute transportation problems,however,the travel satisfaction of its residents has always been among the highest in the world.Using the national metro travel data collected by the automatic fare collection system in Singapore in December 2016,this paper proposes a time series model of inbound passenger flow prediction that considers outliers and an outbound passenger flow prediction regression model that considers spatial characteristics based on the statistical analysis of the overall passenger flow characteristics.On the basis of the existing classic prediction model,the innovative model reduces the influence of outliers on inbound passenger flow prediction,and improves the accuracy of outbound passenger flow prediction through the addition of spatial features.The main work are as follows:1.Visualize the passenger flow entry and exit data collected by the automatic fare collection system(AFC data)in Singapore in December 2016,and abstract the general pattern of subway travel for Singapore residents.The analysis of inbound and outbound passenger flow of the subway station is carried out from the two dimensions of time and space.Summarize the concrete manifestations of the unbalanced temporal and spatial distribution;2.In the study of the inbound passenger flow forecasting problem,the influence of impulsive outliers on time series forecasting was noticed.Aiming at the double seasonality of inbound passenger flow,the Holt-Winter model that has been maturely used is improved,and a time series forecasting method of inbound passenger flow that can effectively reduce the influence of outliers is proposed;3.Considering the strong correlation between inbound and outbound passenger flow when predicting outbound passenger flow,a "double Top K correlation station spatial feature extraction algorithm" is proposed.The extracted spatial characteristics are used for outbound passenger flow prediction.Construct an outbound passenger flow regression prediction model considering the spatial characteristics;4.Empirical analysis.The comparison between the inbound and outbound innovative models and the classic models on the actual data set are conducted.The evaluation result is used to verify the value of considering the outliers and spatial characteristics when improving the accuracy of passenger flow prediction.