Application Of Improved K-Nearest Neighbor Algorithm In Urban Rail Transit Passenger Flow Forecasting

With the rapid economy development,the problem of road traffic congestion is becoming more and more serious.Urban rail transit has the advantages of strong carrying capacity,accurate arrival rate,energy saving and environmental protection.Therefore,people choose urban rail transit as the primary mode of travel,which directly leads to the rapid growth of passenger flow.As an important part of the Intelligent Transportation System,passenger flow forecasting plays an important role in helping the operation department to make good dispatch,cooperate with traffic resources and avoid large-scale passenger congestion,and then becomes an important means to alleviate the overload of passenger flow on subway lines.As a method of passenger flow forecasting,the K-Nearest Neighbor Algorithm is widely used in passenger flow forecasting because it does not need to assume data,is not sensitive to wrong data and has high forecasting accuracy.However,there are some problems in traditional K-Nearest Neighbor Algorithm which lead to excessive forecasting error:inadequate preprocessing of original historical data,unscientific selection of state vectors,slow query speed in pattern matching,and the use of constant K-Nearest Neighbor number.K-Nearest Neighbor Algorithm also has no error feedback and can not adjust the algorithm.In this thesis,the problems mentioned above are improved.Taking Beijing subway passenger flow data as an example,the passenger flow data are obtained from the Automatic Fare Collection System.Due to the similarity between urban road traffic and urban rail transit passenger flow,the urban road passenger flow forecasting method is introduced into the urban rail transit passenger flow forecasting,and the forecasting algorithm is improved in the following three aspects.(1)Improve the processing of historical data.The threshold value method is used to screen the historical passenger flow data for the outliers.The threshold value is selected by the 3? criterion,the wrong value is corrected by the historical average weighting method,and the weather data is added to the passenger flow data.Finally,the Principal Component Analysis method is selected to select the component of the state vector.(2)Improve the historical database.The historical data processed in the previous process are clustered and analyzed to generate clustering centers and clustering data clusters,which are respectively stored in two sub-libraries.The hash function is used to match the current passenger flow status with the clustering centers in the historical database in order to meet the real-time performance of the algorithm.(3)Improve the forecasting algorithm.Using the dynamic K-value method,different neighboring numbers K are selected for different data clusters;error feedback is added to the similarity measure criterion to correct the errors.Finally,this thesis selects the passenger flow data of Fuxingmen Station of Subway Line 2 and Tiangongyuan Station of Subway Line 4 on weekdays and holidays as an example to analyze,obtains the experimental results and demonstrates that the improved algorithm has better performance with the traditional K-Nearest Neighbor Algorithm and other forecasting algorithms.The forecasting accuracy has been greatly improved to a certain extent.At the same time,some future research directions have been proposed for this research.There are 74 references,41 Chinese and 43 English references,44 figures and 14 tables.