Research On Short-term Bus Passenger Flow Prediction Model Based On GCN-LSTM

With the rapid development of China’s economy and the continuous advancement of urbanization,urban road construction has been unable to meet the growing traffic demand,traffic congestion,air pollution,urban road damage and other problems are more and more serious.In this context,the national transport department has formulated a policy of giving priority to the development of public transport,encouraging urban residents to travel by public transport in order to ease the pressure of urban transport.Short-time bus passenger flow prediction research is an important direction in the field of urban transportation,which can improve the efficiency of urban bus operation,help optimize bus route arrangement,provide convenience for urban residents’ bus travel,and greatly promote the development of urban public transportation.Based on the existing prediction at home and abroad on the basis of this thesis summarizes the research methods,aiming at the shortcomings of the existing in current prediction methods,combined with the depth of the neural network can dig the characteristics of the inherent characteristics and regularity in huge amounts of data,proposed figure convolution based neural network（GCN）and length of neural network combination forecast model（LSTM）,Further improve the prediction accuracy of short-time bus passenger flow.The main work is as follows:（1）Data processing and feature analysis.Data cleaning and normalized processing were carried out on the one-card swipe data,vehicle GPS data and weather data of Guangzhou public transport,and the identification of bus boarding and boarding stations was carried out as the data basis for the subsequent analysis of passenger flow characteristics.Combined with the data analysis method,the temporal and spatial variation rules of bus passenger flow data were analyzed from the two dimensions of time and space.Then,10 min was selected as the time granularity to divide the time interval and construct the characteristics of passenger flow.（2）Construct a short-time bus passenger flow prediction model based on LSTM.From the bus passenger flow time dimension change rule to predict the future short-term passenger flow,the historical bus passenger flow data mapping,for one dimensional time series data,and is divided into training set and test set,through parameter Settings,the iterative training for model tuning,to guangzhou bus working days and working days for short-term passenger flow forecast,and the results are visual display.（3）Construct a short-term bus passenger flow prediction model based on GCN-LSTM.From two dimensions of time and space change rule to predict short-term passenger flow,through the length of the time characteristics of neural network to extract the transit passenger flow data,using graph mining convolutional neural network of bus lines network spatial dependence,on the basis of the bus to guangzhou working days and working days for short-term passenger flow forecast,and visual display of results.Finally,compared with common passenger flow prediction models,the average absolute percentage error（MAPE）of GCN-LSTM model was found to be the lowest,and the errors of working days and non-working days were 10.1252% and 12.1432%,respectively,which proved that the model proposed in this thesis had better prediction effect.