Research On Traffic Travel Time Prediction Based On Deep Learning

In recent years,Chinese economy has been developing rapidly,and the level of urban modernization has been deepening.However,the rapid increase of the number of traffic vehicles has brought enormous pressure to the operation of urban traffic.Accurate travel time estimation of public transportation such as taxis and buses,not only can help people to make reasonable planning of the itinerary and save time,but also alleviate traffic congestion and avoid the waste of manpower and energy.Thus,it provides an important reference for urban planning and construction.Traffic travel time estimation has become one of the hot issues in the field of intelligent transportation.However,traditional research methods cannot fully extract the temporal and spatial characteristics of each trajectory path,and ignore the influence of other external attributes such as weather and date,thus fail to achieve accurate time prediction effect.Therefore,this paper takes the accurate estimation of traffic travel time as a starting point,and proposes two travel time estimation models for taxis and buses respectively.The models use deep learning technology to fully explore the deep-level spatial-temporal characteristics of the traffic trajectory data,model and analyze the spatial-temporal dependence characteristics.Furthermore,the influence of other factors such as weather is fully considered.The main research contents and innovations of this paper are as follows:(1)Considering both the waiting time for a taxi and the traveling time when people take a taxi simultaneously,we propose a taxi travel time estimation model based on endto-end training.First of all,by fully considering the taxi's passenger loading status,the acquired taxi GPS point data is cleaned and processed to match the driving trajectory from the unloading state to the loading state of the taxi.The model is mainly composed of external factors,temporal and spatial convolution mechanism and multi-task learning mechanism.The model uses a combination of convolutional neural network and temporal convolutional network to extract the spatial and temporal characteristics of sub-paths.In the multi-task learning mechanism part,it uses the combination of sub-path prediction and the overall path to avoid the influence of error accumulation and data sparsity.(2)We propose an estimation model for bus travel time based on multi-source data.The model is based on the taxi travel time estimation model,and proposes to use the mixed data set to estimate the travel time of buses.It is fully considering the correlation between the mixed trajectories of cars,buses and motorcycles and the spatial-temporal characteristics of trajectory paths.In addition,the model uses the method of dilated convolution and gated recurrent unit to extract the temporal correlation between sub-paths.Through the fusion processing of the features of each component,the accurate estimation of the travel time is realized.Through a large number of comparative experiments on real datasets,it is verified that the error of the model proposed in this paper is reduced and the accuracy of the estimation is improved,when compared with the existing models in terms of average absolute error,root mean square error,and average absolute percentage error.Therefore,our model can be better used to solve the problem of traffic travel time estimation.In addition,due to the applicability of the model,it can also be extended to other time series data estimation problems.