Research On Bus Travel Time And Service Time Prediction Based On Multi-source Data

With the rapid urbanization,traffic congestion and other problems are becoming increasingly serious.Public transport travel is considered an effective way to address the issue,but suffers from problems such as lack of attractiveness,where the bus running stability is one of the important points that affect the appeal of public transportation.Intelligent public transport data collection can provide real-time transit information,which contributes to the reliability of public transport systems.In this context,based on bus GPS data and IC card data,the thesis conducts feature analysis,influence factor mining and prediction model construction for bus travel time and bus service time,so as to provide theoretical basis and data foundation for bus arrival time prediction and information release.This thesis is conducted on the following studies with bus vehicles:(1)The data was cleaned according to the causes of data errors,and GPS data was matched to stations using distance thresholds.Then,the bus station service time calculation method was designed based on speed,time,and mileage,and information such as inter-stop travel time and low-speed driving rate was extracted.Finally,the cleaned IC card data was preprocessed to determine information such as station passenger flow.(2)By analyzing the bus operation pattern throughout the day,the variation characteristics of the bus operation time of different schedules were determined.The distribution characteristics of inter-stop travel time and station service time data were studied,and the conclusion that inter-stop travel time and station service time both obeyed a lognormal distribution was reached.On this basis,the correlation coefficients were used to analyze their spatial and temporal correlation,and to identify the variables that are correlated with the bus travel time and service time of the current schedule.The travel time instability coefficient was constructed to determine a critical value of 0.2 for trip time stability and instability.(3)The dynamic and static factors affecting bus travel times were initially selected in terms of roads,traffic and the environment,and six significant influencing factors were identified through the establishment of a generalized linear mixed model: generalized,number of footpaths,bus lane,previous bus low speed rate,working day and time period.Then,a WOA-ELM bus travel time prediction model was constructed.The MAPE of the forecasting model is 6.50% and 6.07% and the MSE is 122.54 and 149.73 for weekdays and non-working days,respectively,which was the highest prediction accuracy compared to the GWO-ELM and PSO-ELM models.(4)The explanatory variables were initially selected from stations,roads,environment and traffic,and 14 explanatory variables were identified after correlation tests.Then,the BO-XGBoost bus service time prediction model was constructed.The MAPE and MSE of the model were 22.80% and 41.91 respectively,which had better prediction performance compared with mainstream machine learning models.The model was also interpreted in conjunction with the SHAP framework to explain the impact of changes in explanatory variables on bus service times.