EMD-based BP Neural Network For Forecasting Passenger Flow Of Public Transportation

With the development of city,problems such as traffic congestion have been emerging,followed by environmental pollution and traveling inconvenience.The public transportation is playing a more and more important part under this situation.It is crucial to analyze and predict short-time passenger flow in order to improve the efficiency of public transport operations,which can provide decision-makers with valuable information so as to develop a reasonable and efficient scheduling plan.In terms of collecting massive passenger flow data,traditional collecting methods are no longer suitable while intelligent transportation system makes it possible.The basic information such as boarding site and passenger flow are obtained through combining IC card data and GPS data.Based on passenger flow data,we can conclude the overall trend and major characteristics of passenger flow in weekday and weekend,within one day and one month.In addition,peak interval division leads to better understanding of passenger flow.Predicting nonlinear and non-stationary passenger flow data is beyond previous statistical approaches.On basis of BP neural network,one of the most classical artificial neural networks,this paper proposed improved hybrid models.One model is the combination of particle swarm optimization algorithm with BP model,with the purpose of optimizing connection weights and threshold values of the network.The other one is the combination of empirical mode decomposition method and BP model,which aims to reduce the non-stationarity of original data.Final results indicate that hybrid models can improve the prediction accuracy and performance.