Highway Traffic Flow Prediction,Simulation And Application

Short-term traffic flow prediction and traffic flow modeling are critical components of the decision support system in freeway operation and management.They are important to improve the performance of freeway management.This study mainly focuses on short-term traffic flow prediction and traffic flow modeling.Through analyzing the deficiency of existing studies,targeted measures are presented.A novel short-term traffic flow prediction method and new traffic flow model are proposed.Typical applications in freeway emergency management verify the effectiveness of proposed methods.The main contribution of the study are summarized as follows:(i).The freeway short-term traffic flow prediction method includes two contents:(a)Two popular used nonparametric prediction methods which are nonparametric regression and neural network in short-term traffic flow prediction are compared in the requirement of data size and robustness for traffic flow fluctuation.It shows that neural network has a stronger requirement in data size as well as more robust than nonparametric regression.(b)Inspired by the feature selection in machine learning and pattern recognition,a mixed feature selection method is integrated into nonparametric regression to predict short-term traffic flow.The method firstly employs Relief F algorithm to select candidate attribute features subset.Then,it adopts genetic algorithm and delta-test as assessment criteria to further refine the above candidate subset to form optimal or suboptimal state vector.The experiment shows that state vector,which derived from the proposed method,can reduce estimated error significantly both in nonparametric regression and neural network under complicated road topology.Meanwhile,it can provide more accurate prediction under traffic fluctuation.(ii).A more realistic freeway microscopic traffic simulation model,which performs better in terms of actual data,is proposed.Based on the last cellular automaton model for traffic flow,the proposed simulation model takes safe speed,logic equation of stochastic probability and small perturbation of cars at low speed into account.The simulation result shows that,the modified model can simulate traffic flow metastable characteristic,spatiotemporal diagrams,phase transition and concavity evolution of traffic oscillations better.In addition,it can reproduce the time sequence of speed detected from sensors or floating cars.Therefore,the proposed model has many advantages over current microscopic traffic simulation models.(iii).A freeway patrol vehicle scheduling approach based on the result of traffic flow prediction is presented.By analyzing the relationship between freeway traffic accidents and relevant external factors,the correlation coefficient between risk weight and predicting data can be obtained from panel data model.Thereafter,the dynamic weight of the probability of car accident on each link of the road network can be calculated via real-time predicting data.Finally,freeway patrol vehicle scheduling model is built.Tabu search algorithm is adopted to solve the model.The experiment result shows that it can improve the efficiency and effectiveness of patrol and rescue process.This research will help improve the operation and management of freeway,providing instructions for daily management of freeway congestion and emergency management under special situations.