Short-term Forecasting Using A Hybrid Model Based On Two-layer Decomposition Technique And Kernel-based Extreme Learning Machine

With the accelerating pace of urban rail transit construction in China,the trend of network operation of urban rail transit system is further intensified and the mode of operating organization is becoming more and more complicated.Therefore,it is an urgent problem to be solved that how to improve the efficiency of transportation,improve service levels and ensure the safety of operation in network operation.As an important part of short-term passenger flow forecasting,seeking a more scientific and reasonable forecasting model is not only helpful to grasp the dynamic changes of passenger flow,make a short-term train operation adjustment plan,but also helpful for the dynamic adjustment of the station management strategy for realizing scientific,efficient and safe operation of urban rail transit.It is of great significance to study the short-term passenger flow forecast of urban rail transit.Firstly,the effective reduction of outliers were realized after the method of identifying and correcting the abnormal value of station traffic in urban rail transit station putted forward based on the defects in the process of AFC uploading.Then the space-time distribution characteristics of urban rail transit was further analyzed.Secondly,according to the non-linear and non-stationary characteristics of short-term passenger flow in urban rail transit,t the ensemble empirical mode decomposition was used for stationary processing of short-term passenger flow,then,the sample entropy was applied to evaluate the predictability of the decomposed components.variational mode decomposition was applied to further decompose the high-frequency components generated by CEEMDAN into a number of modes in order to reduce the randomness of short-term metro passenger flow.Then the paper proposes a short-term passenger flow forecasting model based on kernel-based extreme learning machine optimized by a hybrid algorithm composed of simulated annealing and particle swarm optimization in order to improve the forecasting accuracy.The forecasting accuracy an reliability of the proposed model is investigated with short-term passenger form Xizhimen station of Beijing subway.The results demonstrate the proposed model have a better performance than the other considered models over horizons of one-step,multi-step ahead forecasting,which shows the research of short-term passenger forecasting has a constructive significance to the development of rail transit planning in the future.