Methodology for generating extreme traffic demand patterns with optimization techniques

Abstract

Traffic scenario generation can aid analyzing and evaluating the performance of transportation systems under different traffic conditions. Unfortunately, it has not drawn much attention in practical applications. No standard scenario generation approach can be found so far, other than manual generation, random generation or exhaustive generation, which are neither accurate nor efficient. Given its importance, we propose to borrow ideas from optimization techniques to solve the optimization problems in generating representative and/or extreme traffic scenarios. In this paper, five optimization methods, i.e., Linear Programming (LP), Dynamic Programming (DP), Greedy Algorithm (GR), Genetic Algorithm (GA), and Tabu Search (TS), are introduced and implemented to generate extreme traffic demand patterns. The accuracy and efficiency of these methods are explored with a case study, in which the best and worst traffic demand patterns are searched for an abstract grid network. It is found in the case study that when the problem uses an independent traffic assignment model (e.g., shortest path assignment), LP and DP are the most accurate methods. GR is the most efficient method. GA presents certain accuracy when searching for best demand patterns, while TS is more accurate when searching for worst demand patterns. Moreover, GR, GA, and TS do not require linear constraints and objective functions, thus they could also work with other dependent traffic assignment models if computational times become affordable. Considering the overall performance (accuracy, efficiency, and constrains in application), GR is recommended as the best methods for generating extreme demand patterns.