Research On Bus Passenger Flows Analysis And Short-term Forecasting Based On Multi-source Data

By comprehensively analyzing the basic experience and lessons of domestic and international urban road traffic development,we can draw an indisputable fact that the vigorous development of public transport and improving the utilization rate of public transport are the most basic and effective way to alleviate urban traffic congestion.Only when an efficient,fast,safe and punctual intelligent public transportation system is established,residents will give priority to public transportation.The construction of an intelligent bus system is inseparable from a detailed analysis of the characteristics of bus passenger flow,and even more accurate short-term passenger flow prediction.With the construction of modern informatization,the accumulation of massive bus data resources,efficient big data mining methods,and deep learning algorithms,all provide strong data and technical support for the research of this topic.Based on a comprehensive,multi-angle and in-depth investigation and survey of the current situation and development experience of the bus industry in Guangzhou,Shenzhen,Xiamen and other cities where my country's bus construction is relatively developed,this article takes the bus operation of Hohhot as the entry point and research object.First,pre-processed multi-source data such as public transportation IC card data,vehicle GPS data,weather data,line station data,etc.,to provide a "clean and clear" data foundation for subsequent comprehensive and objective research.After that,using data mining technology,the multi-dimensional(temporal,spatial,weather,crowd)distribution features of the bus passenger flow hidden in the multi-source bus data information were extracted.Next,taking the passenger flow of 27 bus lines as the prediction object and 15 minutes as the prediction time granularity,the three short-term passenger flow prediction models are constructed using the three algorithms Gradient Boosting Decision Tree(GBDT),Long Short-Term Memory(LSTM)and Extreme Learning Machine(ELM).In the process of model building,not only the parameters of the algorithm itself,but also the key factors influencing the short-term bus passenger flow.The main result orientation is to deal with the global adaptability,training timeliness and result accuracy of multi eigenvector prediction,three different network structure prediction models are comprehensively compared and analyzed.The results show that GBDT and ELM have strong global processing ability,and have high accuracy and superiority in processing short-term bus passenger flow prediction with multi-feature input.Finally,combined with the actual situation of bus operation,the application value of the distribution characteristics and short-term forecasting results of public transport in realizing the scientific planning of public transport system,reasonable scheduling,efficient operation,improving passenger travel experience,and improving the utilization rate of public transport are analyzed.