Research On Subway Short-term Passenger Volume Forecasting Based On Travel Mode And Neural Network

Building rail transit system is an effective way to alleviate the traffic pressure.Accurate prediction of short-term passenger flow can provide the basis for intelligent dispatching,site current limiting and passenger evacuation scheme.Due to the nonlinear and time-varying characteristics of short term passenger flow,neural network is chosen as the prediction model in this paper.Since the performance of neural network is largely dependent on the initial parameters of the model,an improved firefly algorithm is proposed to optimize the initial parameters of neural network.The main work we have done is as follows:(1)We analyze the characteristics of subway passenger flow base on transport IC card data of Shanghai: Because of the differences in passenger flow characteristics within a week,we use the hierarchical clustering method to cluster passenger flow and analyze the passenger characteristics of Friday and the day before long holiday according to the annual report of Shanghai comprehensive traffic,on this basis,we further refine the clustering results;We calculate the Spearman correlation coefficient between passenger volume of the time slice to be predicted and passenger volume of the historical time slice;We design 10 kinds of travel modes base on related theory of the passenger travel chain,on this basis,the definition of“commute” in this paper is proposed.We calculate the commute passenger volume of each time slices for all subway stations on the Hadoop platform.In addition,we verify that the commute passenger flow separated from the transport IC card data has temporal-spatial stability over a period of time.(2)We analyze the impact of rainfall and air quality index on short-term passenger flow based on the relevant data provided by Shanghai Meteorological Bureau and Environmental Protection Bureau.(3)We choose the passenger flow sequence with high time correlation as the input of prediction model,and compare the prediction accuracy of BP network and Elman network.Then,we further narrow the interval of the optimal input dimension of the passenger flow sequence and select the Elman network which is more suitable to time-variability as the prediction model based on the experimental results.By combining commuting and short-term passenger flow forecasting,we verify that the commutation factors proposed in this paper can greatly improve the prediction accuracy.Meanwhile,we choose the best input for the prediction model to carry out the following experiment.(4)We introduce the meta-heuristic optimization algorithm and analyze the theory,process,advantages and disadvantages of the firefly optimization algorithm(FA)in detail.Some improvements are proposed for its shortcomings: the chaotic mechanism and the“catfish effect” are added to the algorithm to improve its global search ability;the Levy flight is introduced into the algorithm to enhance its local exploration ability;each firefly is given adaptive step size to improve the optimization precision of the algorithm.The effectiveness of the improved FA algorithm is verified by comparing the convergence speed and the accuracy of the optimization algorithm;the validity of the improved model based on the improved FA algorithm is verified by comparing the prediction performance of different models.