Research On Public Passenger Flow Missing Data Filling And Short-term' Passenger Flow Forecast Methods

Due to the rapid development of science and technology,data collection and data storage technology have made rapid progress,which has produced a large number of traffic flow data,but there are also a lot of missing or damaged data.The accurate and timely transmission of bus passenger flow information is the basis of intelligent bus scheduling and plays an important role in intelligent bus scheduling and control.Therefore,in order to provide better data basis for bus dispatching and better guide intelligent bus dispatching,it is necessary to study the methods of missing data filling.In this paper,we want to improve the correct rate of the missing data of bus passenger flow and the accuracy of predicting bus passenger flow.This paper uses machine learning algorithm to process bus traffic data,based on field-stack noise since the encoder model fill for lack of public transport pas.senger flow data,and put forward the LSTM based network and decision tree model for short-term transit passenger flow forecast.In order to better introduce the missing data filling model of bus passenger flow,this paper first introduces AE,DAE,DSAE algorithm,and then proposes the method of auto-data filling based on neighborhood-de-noising stacked auto-encoder.This method firstly improves the given range of neighborhood,then finds the missing data of.bus passenger flow to fill in the highly correlated passenger flow data according to neighborhood knowledge,and USES neighborhood matrix as the input value of DSAE network.Finally use public transport passenger flow data provided by the city bus group,validation based on NN-DSAE the lack of public transport passenger flow data packing method of performance under different data loss rate,at the same time,the algorithm and the historical average method and the traditional neural network algorithm to fill the result comparison,the experimental results show that the lack of public transport passenger flow data based on NN-DSAE fill fill effect is better than the other two algorithm of the model.Short transit passenger flow forecast model is proposed in this paper a combined model,first after the experiment than found inside the best wavelet analysis wavelet thresholding function,and then select threshold according to the experiment,this article selects the db3 three layer wavelet decomposition function,after missing passenger flow data fill the bus passenger flow data is decomposed into uniform part and random part.,bus passenger flow uniform part use LSTM network to forecast,and use the city bus group is provided by the passenger flow data to verify the accuracy of the algorithm,and ARIMA,and RNN SNN,regression tree algorithm,to illustrate the advantages of the proposed algorithm,the experimental results show that LSTM network prediction of passenger flow data is much more stability and also higher prediction accuracy.Then the random part of bus passenger flow data is used as the adjustment error term of short-term passenger flow prediction,and the decision tree method is used to predict.Finally,the combined results of the uniform part of passenger flow predicted by LSTM network and the random part of passenger flow predicted by decision tree algorithm are taken as the final passenger flow predicted,and the results are compared with other individual LSTM or decision-tree in this paper has better effect.After the proposed NN-DSAE passenger fill algorithm to fill out passenger flow data,on the basis of the LSTM neural network and decision tree model is applied to forecast the short bus passenger flow data,through the two steps of processing,for bus group provides a more accurate prediction model,for the goal of a intelligent bus dispatch bus group has laid a solid foundation,further promote urban construction and development of wisdom.