Deep Network Model For Travel Time Prediction Based On Travel Planning Data

The development of smart cities presents new challenge to more convenient and intelligent transportation,while the advent of big data era promotes the sharing of travel data,which provides powerful data support for the development of smart travel.As an important part of intelligent travel,accurate travel time prediction is very important and has always been a hot issue in the field of transportation travel.The existing related research models are passive statistical model,that is,the characteristics of real historical data are counted to achieve the future prediction passively.The lack of future travel demand information limits the performance of the model,which leads to such models cannot always maintain the accuracy of prediction.Based on this,this paper proposes a deep neural network model based on travel planning data,which can actively capture the future demand information and “deduce” the future travel time by exploiting the data of future planned traffic volume.Specifically,we collect the travel plan information released by users on the navigation before travel,and convert it into the same amount of planned traffic data in the future.The idea of open data sharing advocated in the era of big data provides a higher possibility for the complete acquisition of travel plan information.As the future planned data,this data and corresponding historical data will be input into the model to realize the prediction of future travel time.Based on above research ideas,the main research works of this project is as follows:(1)An estimation algorithm is proposed to estimate the future planned traffic volume from the travel plan data.The algorithm can transform the travel plan information extracted from the trajectory data into the equivalent planned traffic volume at the future time.The experiment proves the accuracy of the estimation algorithm and the feasibility of using the data as the prediction input of the travel time model.(2)A deep neural network model considering future planned data is proposed for travel time prediction.The model includes three modules,namely,feature input module,feature learning module and travel time prediction module.Firstly,tensor matrices were constructed for future plan data and corresponding historical data as model input.Then the feature learning module introduces the residuals network to design different residuals convolution modules to learn the time and space dimension features of the two input tensors.Finally,the prediction module introduces the GRU network and takes the learned historical features as the input of hidden layer state in the past moments,and the learned future plan features as the input of the sequence that needs to be predicted in the future moments,so as to realize the travel time prediction considering the future planned features.(3)The track data of Chengdu road network published by Didi data sharing platform is used as experimental data to verify the performance of the model through a large number of experiments: the input time of multiple historical data and the forecast time of future planning data are set respectively to train and verify the accuracy and stability of the model;Considering a variety of traffic conditions,the test set is additionally divided into frequent congestion data set and occasional congestion data set,and the overall performance of the model and its foresight in predicting congestion are verified by comparing with other benchmark models in three traffic conditions;Finally,the model is compared with the model without considering travel planning data of the same structure to prove the additional performance of the model brought by travel planning data.The experimental results show that compared with the traditional travel time prediction model based on historical data,the neural network model proposed in this paper is better than other benchmark models in various traffic conditions,especially in the prediction of occasional congestion.