Research On Passenger Flow Prediction And Early Warning Of Urban Public Transport Based On Multi-source Data Fusion

A public transportation system is a vital component of the urban transportation system,with its efficiency having a direct and major impact on daily travel to residents.Due to the absence of energy and land resources,developing urban public transport is one of the most effective ways to solve residents’ travel problems and alleviate urban traffic congestion.In China,many cities have invested substantial capital and resources for improving infrastructure,especially in promoting bus systems,but the rising demand for urbanization could still not be satisfied.In recent years,with urban rail transit systems advancing considerably and under the impact of many external factors,the prevalence of bus systems among passengers has gradually declined and the bus passenger flow in Beijing has shown a downward trend.Therefore,it is urgent to improve the quality of bus service.At the same time,however,techniques that acquire real-time data from IC cards in bus systems have developed markedly,which allows reflecting immediate information on the passenger flow.Therefore,it brings benefits for analyzing the bus passenger flow based on fused multi-source data.In this research,it should be noted that the term ‘bus’ used refers exclusively to regular buses within the city limits.Based on the mentioned above,for improving the efficiency of management and the service quality,and promoting the mass appeal to passengers,the research data in this research is drawn from multiple sources: IC card data,bus route data,bus stop data,bus daily passenger flow,weather condition,holiday information,Position of Interest(POI)data,and COVID-19 data.In this respect,it allows analyzing the bus passenger flow from multiple dimensions,such as spatial dimensions(whole bus networks and bus stops)and temporal dimensions(short-term and mid-term).The research methodology encompasses analyzing,predicting,and early warning,which aims to provide a scientific approach for bus companies,planning departments,and supervisory departments,and to help them make decisions.The overall structure of the research takes the form of four parts and has been organized in the following ways:(1)Spatial-temporal characteristics analysis of bus passenger flow.In bus systems,problems still exist where bus capacity could not match the passenger flow,which leads to low quality of bus service,waste of bus resources,and carbon emissions increase.Spatial-temporal characteristics analysis is to help predict the convergence degree of passenger flow for bus companies,which allows them to understand the true command of travel,sustain the balance between traffic demand and capacity,and increase passenger flow sharing rate.To solve the problems mentioned above,spatial-temporal characteristics analysis has become one of the most significant approaches.Numerous studies have attempted to investigate bus passenger flow in spatial-temporal characteristics analysis,while insufficient explanatory analysis on the result.To mine and analyze the bus passenger flow from spatial and short-term dimensions,this research has designed a clustering model to analyze the features of bus passenger flow,with the model combining a k-means clustering algorithm based on time series similarity calculation and a POI verification model(TSSC-KM-POI).The model can be used to cluster the bus passenger flow time series and explain the results.In the case study,the model mines and analyzes the distribution regularity of passenger flow in the spatial-temporal dimension,which provides a scientific decision-making foundation for optimizing the bus lines and bus stop distribution.(2)Mid-term passenger flow prediction around the whole bus network.Developing a mid-term plan for administrators in a bus system is concerned as a necessary and important work while preparing the plan depends heavily on a more accurate prediction of the passenger flow.Since the mid-term passenger flow has several characteristics(such as time-dependence,periodicity,complexity,and diversity),bus companies and planning departments fail to apply a proper approach to predict mid-term passenger flow.Under this circumstance,this research proposes a hybrid predicting model—an improved STL(Seasonal-Trend decomposition procedure based on Locally weighted regression)based Long Short-Term Memory(LSTM)prediction model(ISTL-LSTM)to predict bus mid-term passenger flow.Considering the characteristics of mid-term bus passenger flow,the model uses the STL decomposition method to indicate the trend and periodicity of bus passenger flow time series and applies the LSTM method to capture the mid-term dependence of future passenger flow and historical passenger flow by memorizing historical information,and introduces multiple factors that affect the midterm bus passenger flow to reflect its complexity and diversity.The results suggest that the prediction performance of the model is improved compared with the existing benchmark methods,which could provide data and strategy support for bus companies and planning departments.(3)Short-term passenger flow prediction of bus stops.In bus operation management,bus companies arrange schedules mainly depending on manual operation,which is difficult to make real-time schedules according to the actual dynamic change of passenger flow.When setting real-time schedules and allocating drivers and buses,bus operation could reference the prediction result of short-term passenger flow.However,due to the high nonlinearity and instability of short-term bus passenger flow,traditional prediction methods failed to solve these problems effectively and achieved poor performance in predicting short-term bus passenger flow.Under this circumstance,this research proposes an STL-based Ridge regression Stacked generalization ensemble model(STL-RS).The conclusion indicates that the STL-RS model has better robustness and accuracy,with prediction results providing reference and evidence for scheduling,planning,and management for bus operators.(4)Early warning of abnormal large-scale passenger flow of short-term bus stops.By far,existing research on early warning of short-term bus passenger flow focuses on conducting a direct comparison of the predicted value and the actual value to provide a simple early warning of the future abnormal large-scale passenger flow.In addition,there are several problems in these studies by lack of comprehensive warning mechanism,unclear warning levels and warning limit benchmarks,and lack of systematic and quantitative analysis methods for early warning.Based on passenger flow feature clustering,short-term bus passenger flow prediction data,and multi-source data,this research applies ensemble learning to classify and predict,and proposes an early warning method for short-term and abnormal large-scale passenger flow of bus stops.The experiment results indicate that this method performed well in early warning,which can assist bus companies and supervisory departments to better understand and grasp the trend and influence of the abnormal larger-scale passenger flow in real-time,and can provide them with reference and suggestions for decision-making in an emergency.