A Research On Service Level Passenger Flow Prediction Of Ground Bus Based On Multi-task Deep Learning

In recent years,the number of motor vehicles in Chinese cities has been increasing geometrically,and urban traffic congestion has become the norm.Bus priority is an effective way to deal with urban traffic congestion.Ground bus is the main part of urban public transport,but the problem of low service level and low share rate of urban public transport is widespread.The accurate prediction of bus passenger flow is not only the key to improve the service level and sharing rate of bus,but also the basis of scientific bus scheduling and reasonable planning and adjustment of bus routes.However,the traditional prediction model is unable to capture the complex spatiotemporal correlation of bus passenger flow and the influencing factors of internal and external characteristics of the bus system,which leads to low prediction accuracy.Therefore,it is necessary to build a prediction model based on multi-task deep learning.In this paper,the IC card data,bus GPS data,bus line station data,weather data and POI data are preprocessed.Secondly,through data mining to get the situation of passengers on and off the bus station and transfer station,analyzed the influencing factors of bus passenger flow,analyzed the spatio-temporal distribution characteristics of bus passenger flow and its internal and external influence characteristics.Then,a multi-task deep learning(MDL)prediction framework is proposed,which uses a deep neural network called ARM network embedded in the MDL framework to predict the bus service passenger flow,line-level on-board passenger flow,and line-level on-off passenger flow of each bus line arriving at or passing through each station in a short interval.ARM network combines three modules of attention mechanism,residual block and multi-scale convolution,which can well capture various complex nonlinear spatio-temporal correlations and internal and external influencing factors of the bus system.MDL frameworks can reinforce each other’s ARM predictions for each type of flow,eventually integrating the output to achieve fine-grained service-level predictions.Finally,based on the data set of Chongqing bus operation,the model method constructed in this paper is verified.The results show that the MDL-ARM prediction model proposed in this paper is superior to the other 10 advanced baseline models,and the accuracy of the MDL-ARM prediction model is 22.39% higher than the optimal baseline model.In this paper,a bus service level passenger flow prediction model based on multi-task deep learning--MDL-ARM model is proposed,which can well capture various complex nonlinear spatio-temporal correlations of bus passenger flow and internal and external characteristics of the bus system,and improve the accuracy of passenger flow prediction.This study has a certain complement and improvement to the analysis method of spatiotemporal characteristics of bus passenger flow and the forecasting method of short-time passenger flow of ground bus...