Research On Microscopic Traffic Flow Modeling Based On Data-driven

Car-following model and lane-changing model are widely applied in the fields of microscopic traffic simulation,traffic capacity analysis,Adaptive Cruise Control,automatic driving,traffic safety evaluation and so on.And they are also the core contents of microscopic traffic flow theory.The traditional microscopic traffic flow models have a common problem that they are the mathematical models based on mathematical formula and traffic flow theory,which leads to these models difficult to reflect the driver's the uncertainty and inconsistency of a series of psychological and physiological activities,for instance,feelings,understanding,judgment,decision.Starting from the data itself,this paper carries out the research on microscopic traffic flow modeling based on data driven to compensate for the above shortcomings with the unique advantages of machine learning algorithms,and explores a new research direction of microscopic traffic flow model.Firstly,for car-following behavior,based on the linear combination forecasting,this paper combines the advantages of the controllability to safety factors of car-following model based on dynamics and the strong self-learning capability of car-following model based machine leanring,and improves the objective function of the optimal weighting method to build a combined car-following model.Then,for the decision-making stage of lane-changing behavior,this paper uses the machine learning algorithms(BP neural network,support vector machine and random forest)to establish discretionary lane-changing decision model based on data-driven,and pretreat NGSIM vehicle trajectory data using normalization,Principal Component Analysis and other methods,and then trains and tests the model using the pre-processed data to verify the effectiveness of the model.In addition,the importance of factors affecting decision-making behavior is analyzed by the unique advantages of random forest.Finally,for the execution stage of lane-changing behavior,this paper establishes lane-changing execution model based on BP neural network for the first time,and tries to make up the lack of traditional models by the advantages of the self-learning ability and the nonlinear fitting ability of BP neural network.The results show that the prediction accuracy of the combined model is better than that of Gipps model,and the combined model can control the authenticity and safety of the simulation speed by adjusting the weight of the authenticity and safety of the combination model.The data-driven decision-making model based on machine learning algorithm has high accuracy.In the decision-making stage,driver is more concerned with the distance between the lane-changing vehicle and vehicles on the target lane than relative speed,and relative to the leading vehicle on the target lane,the following car has more impact on driver.The data-driven lane-changing execution model has very high accuracy,and it is feasible and effective to establish the lane-changing execution model by data-driven method.