How Do Vehicles Make Decisions during Implementation Period of Discretionary Lane Change? A Data-Driven Research

Abstract

To investigate and compare the lane changing behavior of passenger cars and heavy vehicles during the implementation period (defined as the interval from the start time to the end time of a lane change maneuver), this study applies the gradient boosting decision tree (GBDT) method to model the lane changing behavior of heavy vehicles and passenger cars, respectively. Results show that the lane change models vary with the vehicle types and lane change directions. Different factors are considered by the drivers of passenger cars and heavy vehicles when implementing lane changes to different directions. Partial dependence plots of GBDT models reveal that the influence of independent variables on lane changing behavior is nonlinear and complicated, which means that the same variable leads to various effects on the lane change decision across different vehicle types and lane change directions. In contrast with other state-of-the-art methods, the proposed method can obtain more accurate results. The findings indicate that it is necessary to build specific lane change models based on vehicle types and lane change directions for microscopic traffic simulators and autonomous vehicles.

A Modified Latent Dirichlet Allocation Topic Approach for Driving Style Exploration Using Large-Scale Ride-Hailing GPS Data

“…One of the personalization measures is to simulate human driving behavior by switching to the appropriate driving style in certain scenarios. In the mixed driving environment, simulating human behavior for AVs can make driving behavior on the road more consistent, leading to safer driving and fewer crashes [2].…”

“…For example, French et al [8] found that dangerous driving styles are associated with serious trafc crashes. Generally, aberrant driving style recognition has several potential applications in practice as follows: (1) helping trafc managers analyze driving behaviors and the contributing factors of road crashes and identify driving risks in a timely manner, (2) enabling AVs to take timely measures to avoid conficts with aberrant-style driving and potentially dangerous vehicles, and (3) assisting trafc managers in taking preventive measures for certain accidents in advance to improve trafc efciency and trafc safety. Terefore, studies on recognizing driving styles and extracting aberrant ones are of great importance.…”

“…Te main contributions of this paper are listed in the following: (1) applied the modifed LDA algorithm with a new encoding method to extract driving behaviors from large-scale trajectory data and build a driving style library containing realistic driving styles for each driver individually; (2) revealed and compared the efects of diferent external factors on driving styles, including working days, peak hour congestions, and weather impact, and explored how the composition of driving styles shifted in diferent driving environments; and (3) identifed abnormal and potentially dangerous driving within numerous styles in various driving environments based on the composition of each style from the driving style library.…”