Developing a Hybrid Framework for Modeling Driving Behavior
Integrating data-driven modeling techniques into traditional, mathematical models
Abstract
We have developed a neural network-based model for replicating the driving behavior in discrete-time microsimulation environments. We extended the scope of our research in recent years by building a general hybrid framework for the integration of data-driven methods into traditional mathematical car-following models. This framework enables traditional models to capture nonlinear and complex phenomena in driving behavior and still remain interpretable. To calibrate the parameters of the hybrid model, a probabilistic model-building genetic algorithm is applied in a simulation-based optimization process. This project led to two publications.
The Hybrid Model
The model consists of two implicit parts: one for the estimation of the reaction delay, and one for the calibration of the car-following behavior of the driver. To further clarify the proposed framework, suppose that a driver is following his/her leader in the traffic stream. At time instant t, according to the traffic conditions (speed, gap, relative speed, etc.), the driver’s intended acceleration is calculated by using an equation-based car-following model (EBCFM). Simultaneously, his/her reaction delay in applying the calculated acceleration is estimated using an artificial neural network model, which is called RT-ANN
The Calibration Process
The parameters of both equation-based and data-driven parts of the model are jointly calibrated in a simulation-based optimization process. An estimation of distribution algorithm (EDA) is used for searching the optimum values of parameters.
Evaluating the Performance of the Model
The figure below presents the scatter plots of relative speed (speed of the leader minus the follower’s) versus acceleration/deceleration for real observations and for the simulation results of the hybrid and a traditional model (GM). As can be seen in the figure, the simulation results of the hybrid model are more consistent with the field observations in comparison to the outputs of the GM model.
