Research On Microscopic Modeling And Simulation Validation Of Lane Changing Behavior Based On Dynamic Acquisition Of Driving Behavior

Engineering practice proves that the first principle of solving traffic problems is investigating and exploring the mechanisms of traffic flow. As one of the primary driving behaviors observed in traffic flow, lane changing(LC) behavior is considered to be the important inducement of traffic safety and traffic jams. At the same time, it gains more and more attention. We can reduce the number of accidents and relieve the congestion effectively through studying LC behavior and developing scientific traffic management and control strategies. Till now, due to the complexity of LC behavior, the accuracy of LC data obtained is lower, and the models presented are lack of integrity and reliability. Therefore, how to extract high accuracy data of LC process, how to describe and model the whole LC process uniformly, how to validate and calibrate the effectiveness of presented model, are the key problems to be solved in this paper. First of all, the existing relevant researches were reviewed, which included field data acquisition and processing, modeling the whole process of LC and the calibration of LC model, and this dissertation also interpreted the research idea in this way. Second, lots of data cleaning approach were proposed to reconstruct the trajectories database based on the error structure of trajectories dataset, and the reconstruction results were evaluated. Again, the execution processes of LC were extracted from the reconstructed database, and the LC characteristics and LC effect were analyzed from micro-level. Then, a cellular automata simulation platform was built, two models were proposed, one was based on the execution of LC vehicle and the other was based on the whole process of LC behavior and the numerical simulation and detailed analysis were made from two-lane system and on-ramp system, simultaneously. Finally, the proposed model was calibrated using the reconstructed data, and the whole dissertation formulated a modeling and simulation framework in the micro-level.The major attained in this dissertation are listed as follows:(1) Based on the analysis of the data error, a multi-step filtering method is put forward for reconstructing trajectories datasets of vehicles. Different target and phases of data cleaning tasks were determined in this method based on the feature of trajectory data. The curve interpolation theory, the signal processing method, the optimization method with kinematics and traffic flow dynamic constraint, time series analysis, support vector machine (SVM), and other means of data cleaning, were combined together to rebuild the vertical position(LocalY), velocity, acceleration, and lateral position(LocaX), besides that the lane line coordinates were gotten. The reconstruction results shown that the reconstructed dataset did not exist too many extremes any more, and seemed more realistic. Then this reconstructed dataset can be used in the analysis of the longitudinal and transverse position, velocity and acceleration directly, and it can meet the demand of LC analysis in the micro-level.(2) Based on reconstructed dataset, the start and end points of LC process from spatial aspect were defined in this research. A set of relativity rules for vehicle spatio-relationship and a set of heuristic rules were put forward for extracting LC process and local relationship of every vehicle. And the rules can be applied to different LC scenarios. The main characteristics and related changes during the LC process were extracted, which included:the probability density function of lane changing duration was followed the logarithmic normal distribution, there was significant differences between the durations of MLC and DLC, the LC duration decreased with the increasing of LC vehicle speed, and LC duration was associated with vehicle type, and so on. In addition, some macroscopic traffic phenomenon which was widely accepted were reproduced in this dissertation, such as metastable state, and wide moving jam, anticipation and relaxation effect. The common assumption for modeling LC in micro-level was verified. The next step in the simulation modeling research was inspired by this step.(3) A cellular automata simulation platform with different traffic compositions, different boundary conditions, and different simulation scenarios was established. The reliability of the platform was verified by reproducing the conclusions of the classic single lane and two-lane cellular automata model. The simulation models presenting actual LC characteristics and decision-making process of drivers were proposed, which were referred to as a cellular automata model considering execution process of LC and a cellular automata model considering the effect of turning light respectively. The two models were evaluated in two-lane traffic system and on-ramp traffic system with homogeneous traffic composition and heterogeneous traffic composition. For the purposes of comparison, a original two-lane cellular automata model was evaluated, too. The observed macroscopic traffic phenomena, such as metastable state, synchronization flow were reproduced by means of numerical simulation. The failure LC and driving effects, such as anticipation, relaxation effect were reproduced by the latter model.(4) A calibration method based on the genetic algorithm for cellular automata model was proposed. The three models were calibrated from the macro level and micro level, respectively, based on reconstructed trajectories database. The results shown that the two new models error were superior to the control model, especially in terms of LC characteristic, two models calibration precision was much higher than the original one. Finally the generalization accuracy of two models using cross validation method was shown in the dissertation. And an organic unity of theoretical modeling and validation was gotten.