Fitness landscapes analysis and adaptive algorithms design for traffic lights optimization on SIALAC benchmark

Abstract

Finding optimal traffic light timings at road intersections is a mandatory step for urban planners wishing to achieve a sustainable mobility in modern cities. Increasing congestion situations constantly require urbanists to enhance traffic fluidity, while limiting pollutant emissions and vehicle consumption to improve inhabitants' welfare. Various mono or multi-objective optimization methods, such as evolutionary algorithms, fuzzy logic algorithms or even particle swarm optimizations, help to reach optimal traffic signal settings. However, those methods are usually designed to tackle very specific transportation configurations. Here, we introduce an extended version of the sialac benchmark, bringing together several real-world-like study cases with various features related to population, working activities, or traffic light devices. We drive a fitness landscape analysis on these various benchmark instances, which helps to improve the design of optimization algorithms for this class of real-world mobility problems. Thereby, we propose a new adaptive optimization algorithm to tackle each scenario of the benchmark.

Review of Traffic Assignment and Future Challenges

“…One approach is to carry out several iterations in which the two problems, i.e., traffic assignment and traffic-light optimization, are dealt with in sequence. In the work presented in [135], the authors use several heuristic strategies [136] to optimize traffic lights while using MatSim [93] for traffic assignment. The authors of [137] suggest a navigation rule based on gene expression programming.…”

“…Furthermore, the solution space exploration is limited [87]. Consequently, several authors propose the use of heuristics and meta-heuristics for dynamic assignment problem resolution, such as simulated annealing [86,90,91], population-based search (e.g., genetic algorithm and teaching-learning-based optimization) [86,[92][93][94][95][96], and ant colonies [97,98]. A noticeable work presented in [99] introduces a new modeling paradigm for DTA and uses multi-agent reinforcement learning to solve DTA.…”

PUBO$$_i$$: A Tunable Benchmark with Variable Importance

“…For example, in [17,18] the authors study a machine learning-enhanced recombination that incorporates an intelligent variable selection method for multi-objective optimization and show that taking the variable importance into consideration can improve the optimization quality. Another example concerns the fitness landscape analysis of landscapes from the SIALAC benchmark, a benchmark for mobility problems [12]. In this benchmark based on a city mobility simulation, variables have different degrees of importance, and a bandit descent heuristic is proposed to take important variables into account.…”

“…One of the peculiarities of real-like instances is that, contrary to instances generated using random-based techniques, they are structured. Thus some variables of the problems play a more important role in terms of fitness contribution, and can be interdependent to several other variables, making them harder to study and understand [12]. Although variable importance of such problems must be studied, existing benchmark generators do not allow users to tune this parameter, which contributes to the lack of instances having important variables.…”

Sparse Surrogate Model for Optimization: Example of the Bus Stops Spacing Problem

“…Parallel approaches benefit from larger computational resources to reduce evaluation time. As the number of evaluated candidate solutions is limited, researchers also define optimization algorithms to increase the convergence rate toward the most promising solutions [24]. Lastly, Surrogate-Assisted Optimization (SAO) builds an algebraic model from evaluated solutions, substituting the original optimization function with the surrogate to guide the search.…”