Short-Term Passenger Flow Forecast Of Urban Rail Transit Stations Based On AFC Data

With the acceleration of urbanization in China and the rapid development of the economy,urban traffic congestion is increasingly serious because urban traffic is affected by factors such as road capacity.In recent years,in order to alleviate traffic congestion and provide better travel experiences to the citizens,more and more cities choose to develop mass transit public transportation systems.More and more domestic cities have chosen to develop rail transit systems due to its large volume,pollution-free,green and energy-saving advantages.Therefore,it is particularly important for rail transit operation companies to accurately predict the passenger flow into stations,in order to develop a more reasonable rail transit operation plan and improve the quality of rail transit services.By summarizing the existing short-term passenger flow forecasting methods in domestic and abroad,aiming at the shortcomings of different forecasting methods,this paper proposes to apply the deep learning method to the field of urban rail transit short-term passenger flow forecasting,and proposes two kinds of deep learning short-term passenger flow forecasting models:PSO-LSTM model and DBN-ELM model.Since it's difficult to determine number of neurons,efficiency of learning,and number of iterations based on LSTM,empirical method and trial-and-error method are always used to calibrate parameters in this model.It leads to a low precision of prediction.This paper proposes to use PSO to optimize parameters in model to increase the precision,thereby establishing a universal neural network for long and short term memory which can be used for short-term passenger flow forecasting of urban rail transit.This paper designs and implements the DBN-ELM prediction model based on the Deep Confidence Network and the Extreme Learning Machine model.The model is based on the unsupervised learning of the Deep Confidence Network and the supervised learning of the Extreme Learning Machine to train the sample data and reconstructs and extracts the data through the underlying Deep Confidence Network,thus providing effective top-level limit learning machine with data expressions.The top-level Extreme Learning Machine fine-tunes the weight of the underlying Deep Confidence Network,making the overall model more suitable for the data change characteristics.At the same time,the PSO algorithm is designed to optimize the parameters of the top-level ELM.Based on the example of passenger flow forecast of Wangfujing Station in Beijing,the validity of the PSO-ESTM model and DBN-ELM model established in this paper for short-term passenger flow forecasting of urban rail transit is verified.By comparing and analyzing the different models and combining the structural characteristics of each model,it is concluded that PSO-ESTM model has better effect on passenger flow prediction with strong time series and DBN-ELM model has better effect on passenger flow prediction with strong randomness and non-linearity.Figures:43;Tables:15;References:53(No count for sub-figures and sub-tables)...