Abstract: With the rapid development of urban rail transit,the short-term passenger flow forecast of the subway is conducive to the operation department to observe the real-time changes in passenger flow and adjust the scheduling strategy.This paper studied the temporal and spatial characteristics of passenger flow.Under the 10-minute granular time slice,there is a periodicity of passenger flow changes,and there are differences in the waveform of passenger flow in space.This paper used agglomerative hierarchical clustering algorithm to analyze the passenger flow of different stations for a week,and obtained the results of passenger flow close to the characteristics of the station.According to the results of classification,correlation analysis was performed on time slices of different types of historical passenger flow,and prediction models based on Support Vector Machine were proposed,regarding time slice sequences with strong correlation as input.Besides,a parameter optimization model for double-population firefly algorithm based on cooperative self-adaptive adjustment was proposed,in which the chaotic attraction was introduced to improve the global search ability,avoiding the initial value being trapped into a local optimum.The adaptive search step length was added to improve the convergence speed and solution accuracy.Compared with other models and optimization algorithms,the proposed model has better prediction accuracy,stability and robustness.

Key words: Chaos, FA, Forecast of passenger flow, SVM, Time series