Research On Short-time Bus Passenger Flows Forecasting Based On Deep Learning

The short-time change rules of bus passenger flows are the basis and premise of all bus operation planning.At present,bus companies generally have problems of short-time passenger flows forecasting ability.Because of the characteristics of the ground bus,the bus passenger flows are more susceptible to external conditions such as weather,which have great randomness and complexity.With the development of urban informatization level,a large amount of public transport IC card data assets have been accumulated,and the deep learning forecasting methods have demonstrated advantages such as high efficiency and strong processing capability in large-scale data analysis,which provides data and technical support for the analysis of bus passenger flows rules and accurate forecasting of short-time bus passenger flows.Based on data mining analysis and deep learning forecasting methods,this paper establishes bus passenger flows analysis system and short-time bus passenger flows forecasting models.Firstly,the single data sets of IC card data,weather data,line data and station data are cleaned,transformed and integrated into a unified multi-dimensional data set as the data source for subsequent analysis of passenger flows rules.Secondly,the data mining method is used to construct the bus passenger flows analysis system.The key influencing factors and the distribution characteristics of bus passenger flows are analyzed from different dimensions,which provides the basis of parameter selection for the establishment of short-time bus passenger flows forecasting model.Then,taking the Xidan area station passenger flows and the 300-express inner line passenger flows as the forecasting objects,and the 5 minutes,10 minutes,15 minutes as forecasting time granularity,using the time series forecasting methods of deep learning-DBN,LSTM,GRU,to construct short-time bus passenger flows forecasting model,and compare the applicability of the three forecasting models to different forecasting objects and different time granularities,and the forecasting performance of the forecasting models of different network structures in short-time bus passenger flow forecasting.The results show that:in the process of short-time bus passenger flows forecasting,the GRU has better forecasting performance than DBN and LSTM,and for the station passenger flows of Xidan area,the optimal forecasting time granularity is 10 minutes;for the line passenger flows of 300-express inner,the optimal forecasting time granularity is 15 minutes.Finally,the application of bus passenger flows distribution characteristics and short-time bus passenger flows forecasting results in bus operation scheduling,urban bus planning,public transportation connections,passenger travel,etc.are analyzed.