| Travel time estimation and travel time prediction is the basis and premise of dynamic path optimization.At present,how to improve the accuracy of travel time estimation and prediction,and the effectiveness of dynamic path optimization is one of the key issues to be solved.The core of solving this problem is reasonable modeling and solving.Therefore,based on the modeling and optimization theory,this paper uses the travel time of an individual vehicle to estimate the travel time of the traffic flow,and predicts the travel time to improve the accuracy of travel time estimation and prediction;then optimizes the dynamic path and improves the efficiency of path optimization.This paper studies the estimation and prediction of travel time and the method of dynamic path optimization.The main work are shown as follows.(1)A method based on the Piecewise Cubic Hermite Interpolation Polynomial was used to estimate the travel time of an individual vehicle in this paper.The GPS data collected on different routes selected as experiment routes.The data processed through data processing,data smoothing,and map matching.The Piecewise Cubic Hermite Interpolation Polynomial method,Piecewise Linear interpolation method and Piecewise Cubic Spline interpolation method were used to estimate the travel time of an individual vehicle under different traffic states(free-flow,transition,and congestion).The comparison of Average Relative Error found that the estimation results obtained by the three methods were similar under the traffic state of free-flow.However,the Piecewise Cubic Hermite Interpolation Polynomial method was obviously superior to the other two methods under the traffic state of transition and congestion.It is found that the Piecewise Cubic Hermite Interpolation Polynomial has the characteristics of better simulating the complex traffic state of the vehicle,which can effectively improve the accuracy of the estimated travel time of an individual vehicle through the road endpoint,and is more suitable for the interpolation of discrete GPS data under complex traffic states.(2)A travel time estimation model based on Random Forests was proposed in this paper.Through feature variables filtering,the final established Random Forests model used 7 influential variables as the model variables,namely mean travel time of floating car tf,traffic state parameter X,density of vehicle Kall,and median travel time of floating car tmenf,speed of vehicle vall,density of floating car Kf and speed of floating car vf.The simulated data were used to verify the model.It is found that different from other machine learning algorithm as black boxes,the Random Forests model not only provided high-precision estimations,but also provided interpretable results through the importance of variables.The result of importance analysis showed that tf,X,Kall,and tmenf were important variables affecting travel time of traffic flow,Vall,Kf and Vf also have some influence.Compared with the Mean Absolute Deviation,it is indicated that the Random Forests model proposed in this paper has higher precision and more abundant variables than the BP(Back Propagation)neural network model and the quadratic polynomial regression model.Therefore,the Random Forests model has more practical application value.(3)A travel time prediction model based on Gradient Boosting Decision Tree(GBDT)was proposed in this paper.11 influential variables were selected as variables of the model.With different prediction horizons(5 min ahead namely 1-step ahead,10 min ahead namely 2-step ahead and 15 min ahead namely 3-step ahead),travel time prediction models based on GBDT were established respectively.The performance of the GBDT model was verified for different cases.The result of importance analysis illustrated that travel time in the current period Ti was the most important influence variable for different prediction horizons.Traffic conditions in the current period had the greatest influence on the predicted travel time.Compared with the MAPE(Mean Absolute Percentage Error),it is showed that the GBDT model proposed in this paper has higher precision than Back Propagation(BP)neural network model and the Support Vector Machine model;especially in multi-step prediction,the advantage is more obvious.Meanwhile,the GBDT model can better explain the influence of variables on the prediction results.Therefore,the GBDT model is more promising in the travel time prediction.(4)The ant colony algorithm was improved to optimize the dynamic path.Through the actual investigation and analysis,the influencing factors of the multi-objective planning model were determined.The ant colony algorithm was improved by using the analytic hierarchy process to transform the path length,travel time,and traffic flow into the comprehensive weight influencing factors.At the same time,the improved ant colony algorithm added directional guidance and dynamic optimization.For different road networks,the effectiveness of the improved ant colony algorithm was verified.Compared with the travel time of optimal path,it is revealed that the improved ant colony algorithm is more accurate than the basic ant colony algorithm and the spatial shortest distance algorithm.Meanwhile,the number of iterations of the improved ant colony algorithm is reduced.Therefore,the improved ant colony algorithm can quickly and efficiently obtain the dynamic optimal path.The research in this paper improves the accuracy of travel time estimation and prediction and improves the efficiency of path optimization.It can provide a theoretical basis and technical support for the dynamic management of road traffic and provide an important basis for the further implementation of Advanced Traveler Information System and Advanced Traffic Management System.It can provide real-time and accurate guidance information for road users.It has a certain practical reference value for improving road traffic conditions,improving management and road use efficiency,and reducing environmental pollution caused by traffic congestion. |
PDF Full Text Request