Research On Driving Behavior Characteristics Based On Internet Of Vehicles

Driving characteristics, which serve as fundamental factors for traffic applications such as vehicle active safety system、cooperative vehicle infrastructure system、intelligent transportation system、traffic safety management and control, has a positive effect to collision warning systems based on vehicular networks technology. Motions during driving process are time-variable and also exhibit characteristics of complexity and stochasticity. Recent developed vehicular networks technology provides a new perspective on this batch of investigation: timeliness state information of each vehicle(for instance, position, velocity, acceleration) in related traffic which can be obtained by using the technologies of vehicular networks. Subsequently, driving characteristics will be extracted from these information and help to improve the precision of predicting vehicles‘ future trajectories and formulate appropriate warning strategy. Meanwhile, researches on driving characteristics act as roots to collision warning systems and require more investigations. The objective of this dissertation is to improve security in the scenario of traffic with vehicular networks, by identifying the interplay and response between ego vehicle and surrounding traffic for different type of driving motions categorized in the paper. This can be reflected in several aspects:In order to curb the instability of the driving characteristics caused by the combination instability effect of driver, vehicle and exotic environment, the driving characteristics in this paper are defined as combination factors that describing both driver and vehicle. That means each participant of traffic in the dissertation, is a combination of vehicle-driver and regarded as a ?black box‘. By analyzing response of surrounding traffic recorded historic state and identifying driving characteristics, vehicle‘s trajectory in a short future can thus be predicted, which is a critical reference for real-time collision warning and can serve as a theoretical fundamental to these systems. The negative effect of complexity physiological and psychological property and the difficulty in corresponding gleaning process for safety application is thus diluted. By using vehicle historical trajectory data, the driving characteristics of each driver-vehicle unit are estimated by three factors of reaction time, minimum spacing and stability, and as a result a self-adaptation cognition model is built in this dissertation.The major factors influencing categorizing drivers, according to NGSIM data(which correspond to traffic data with vehicular network) and Newell‘s car-follow theory, are regarded as reaction time and minimum spacing evaluated in motions responding to extrinsic stimulus. And based on this, we future discriminate drivers as proficient, offensive or normal using clustering algorithm. Besides, comprehensive evaluation of driving behavior is proposed, that is using stability index as another assessing factor to calibrate in asymmetric phenomenon in acceleration and deceleration(hysteresis phenomenon), and subdivide drivers into torpid, prudent, offensive, sensitive, conservative, high-risk or proficient. To assess lateral driving characteristics, lane-changing stability index is used to quantify the changes between before and after lane changing process for lane-changing vehicle and vehicles influenced.Furthermore, a traffic simulation considering driving characteristics is developed to examine this research, and lane changing process in this simulation is subdivided into three phases: intention, requirement judgment and execution, to simulate real lane changing process in traffic. Each of phases during lane-changing is analyzed by using Neural Networks lane-changing model, Bayesian Networks lane-changing judgment model, and lane-changing execution model developed in the dissertation, respectively. The first two models efficiently curb the abnormal motion caused by erratic lane-changing decision-making, while the latter makes the lane-changing motion more realistically.Finally, accuracy of trajectory prediction using calibrated Newell‘s car-following model is examined in the paper, based on the perspective of collision warning application using vehicular network. By comparing to trajectory prediction model based on IDM and OV model, we demonstrated that the proposed model has less deviation in both position and velocity prediction, and the performance of the model which may benefit the theoretical research of real-time collision warning system. We also estimate the reliability in macroscopic traffic by adjusting driving characteristic parameters to reproduce several empirical macroscopic traffic flow such as asymmetric phenomenon in acceleration and deceleration, propagation of stop-and-go wave, anticipation phenomenon before lane-changing and ―regressive effect‖ after lane-changing. Result of macroscopic is positive, and uncover the mechanism of how extinctive factors influence driving characteristics from traffic simulation perspective. It also demonstrates the positive effect in collision warning system construction by integrating the proposed model.