Artificial Neural Network Models for Car Following: Experimental Analysis and Calibration Issues

Abstract

The paper deals with the application of Artificial Neural Networks to model the car following driver's behaviour. The study is based on experimental data collected by several GPS equipped vehicles that follow each other on urban roads. A 'swarm' stochastic evolutionary algorithm has been applied in training phase to improve convergence of a usual error-back propagation algorithm. Validation tests highlight that ANNs provide a quite good approximation of driving patterns and can be suitably implemented in micro-simulation models. More advanced applications to ITS may concern Advanced Driver Assistance Systems and are addressed to conform braking actions to drivers' expectations.

Driving behaviour models enabling the simulation of Advanced Driving Assistance Systems: revisiting the Action Point paradigm

“…Such approaches have more recently been used in ACC design, where studies range from empirical investigations (e.g., Wang et al, 2011, andBoyle, 2013), to modelling exercises (e.g., Bifulco et al, 2008, who proposed implementation of a following model using a machine learning technique), through to large-scale experimental FOT (Field Operational Test) studies, such as those reported in Viti et al (2008). It can be seen, therefore, that studies and models of driver behaviour (in particular during car following) are crucial to making advancements in ADAS (Simonelli et al, 2009), and this is a rich field undergoing rapid development (e.g., Colombaroni and Fusco, 2013, who propose an Artificial Neural Network approach for traffic-simulation oriented car-following models).…”

Estimating and Exploiting the Capacity of Urban Street Networks

“…Experimental activities have been conducted to develop a car-following model based on the Neural Network paradigm using a fleet of few GPSequipped vehicles (Colombaroni et al [25]). This research activity is now addressed to implement a deep learning algorithm to develop a new carfollowing model.…”

Simulation Optimization of Car-Following Models Using Flexible Techniques

“…Reuschel [67] and Pipes [61] introduced the idea of car-following models. Representative microscopic traffic models between the 1950s and the 1970s have been developed by Bifulco et al [14], Chandler et al [17], Colombaroni and Fusco [23], Kometani and Sasaki [44] and Zhang et al [90], Herman et al [40], [27,35,55,79]. Most of them are defined by an acceleration function, which includes the difference of position x i+1 −x i and the difference of speed v i+1 −v i between a vehicle i and its lead vehicle i + 1: the difference of position x i+1 − x i and the difference of speed v i+1 − v i .…”

“…Zhang et al [90] have demonstrated the use of machine learning methods to support the development of data-driven intelligent transportation system. Data-driven approaches have already been used in a fully adaptive cruise control system by [14] or in car-following modeling with artificial neural networks by Colombaroni and Fusco [23]. Finally [59] have performed a data-driven approach based on loess method for speed estimation using Naples data.…”