Data-driven Bus System’s Perception Analysis And Route Optimization

In recent years,with the development of society,urban vehicles become more and more,which is far more than the growth of urban roads.Traffic congestion,traffic safety,environmental pollution have become the three important problems which many cities are facing.Giving priority to the development of public transport,to guide passengers to travel by public transport is an important way to solve this problem.However,the development of public transport systems in many cities in China is not satisfactory.Transfer inconvenience,vehicle crowded,poor quality of service and other issues are also very prominent.In order to meet the diverse,multi-level,personalized travel needs of urban residents,improve the competitiveness of public transport mode of travel,urban public transport system should open customized bus lines outside of the traditional public transport system.At the same time,with the rapid development of "Internet +" concept,bus system has gradually realized the electronic and intelligent.The data we can get has a qualitative improvement in terms of quality,size,type,timeliness.How to analyze and dig out the valuable knowledge hidden in the data for the applications,has become an important research topic in big data analysis and data mining.In this paper,taking Shenzhen as an example,we use urban bus GPS data,passenger tapping IC card data,station location information and custom bus APP purchase data to research from three aspects which are vehicle operating state perception and prediction,traditional bus passenger flow perception and station passenger flow forecast,bus line optimization.The specific contents include: 1.The vehicle operating state perception and prediction: As most of the current urban vehicle GPS data without station time,so we first estimate the vehicle arrival station time.Based on the characteristics of vehicle GPS data reported in Shenzhen area,we designed "interpolation estimation method" to estimate the vehicle arrival time.Based on the vehicle arrival time,the "dynamic weight prediction method" was designed to predict the vehicle arrival time in real-time.Through the actual investigation and verification,our arrival time estimation error within 10 s accounted for more than 90%,arrival time prediction is better than the current Baidu,Gaode map.2.The traditional bus passenger flow perception and station passenger flow forecast: We integrate the passenger tapping IC card data and vehicle arrival time data,design the "travel behavior analysis algorithm" to get the history OD of passengers.We also analyze the time and spatial OD distribution of traditional bus passenger.Based on the historical distribution,a "two step real-time prediction method" based on Kalman filter is designed to predict short-term passenger flow.The results show that our passenger OD estimation accuracy is above 77%,and the station passenger flow prediction is better than the current mainstream prediction algorithm.3.Bus line optimization and effect evaluation: In this paper,a "multi-impact factor passenger flow forecasting algorithm" based on BP neural network is established to predict the passenger flow of different stations during different periods.For those stations which have larger flows,we select the appropriate way to connect them into a line.The line generation algorithm is 500 times faster than the traditional algorithm.At the same time,we use the design of the real design line to verify the passenger flow forecast and the rationality of the line.The results show that our prediction value is more reliable than the traditional method,and the line generation scheme is practicable.