Stochastic Multi-class Traffic Assignment for Autonomous and Regular Vehicles in a Transportation Network

TL;DR

This paper develops a stochastic multi-class traffic assignment model to analyze mixed traffic of autonomous and regular vehicles, introducing new solution algorithms and a path generation method to improve computational efficiency.

Contribution

It introduces three novel solution methods for the SMTA problem and a path generation-assignment algorithm, enhancing analysis of mixed vehicle traffic.

Findings

Modified Wang's algorithm improves convergence speed

Numerical results demonstrated on Nguyen and Sioux Falls networks

Effective path generation for large-scale networks

Abstract

A transition period from regular vehicles (RVs) to autonomous vehicles (AVs) is imperative. This article explores both types of vehicles using a route choice model, formulated as a stochastic multi-class traffic assignment (SMTA) problem. In RVs, cross-nested logit (CNL) models are used since drivers do not have complete information and the unique characteristics of CNL. AVs, however, are considered to behave in a user equilibrium (UE) due to complete information about the network. The main innovation of this article includes the introduction of three solution methods for SMTA. Depending on the size of the network, each method can be used. These methods include solving the nonlinear complementary problem (NCP) with GAMS software, the decomposition-assignment algorithm, and the modified Wang's algorithm. Through the modification of Wang's algorithm, we have increased the convergence speed of Wang's algorithm and shown its numerical results for the Nguyen and Sioux Falls networks. As it is not possible to consider all paths in the traffic assignment, we proposed a creative path generation-assignment (PGA) algorithm. This algorithm generates several attractive paths for each origin-destination (OD), and the modified Wang's algorithm assigns traffic flow. Keywords: Autonomous Vehicles, Stochastic Multi-class Traffic Assignment, Cross-Nested Logit Model