Sincronização de semáforos como um problema de otimização com muitos objetivos

Abstract

Intelligent Transportation Systems (ITS) aim to optimize transportation efficiency and improve safety through the use of advanced technology, encompassing different spaces of application. One of the primary ITS-related areas is traffic management, which uses new concepts in the organization and maintenance of traffic, while seeking to keep good-quality traffic flow, among other aspects. Included in traffic management is traffic light synchronization, which is one of the approaches dealing with the reduction of traffic congestion. Synchronization is achieved when two or more traffic lights are running the same types of traffic patterns so that a vehicle can pass through the synchronized lights without stopping. As a consequence of synchronization, it is possible to optimize a given traffic-related quality, usually the flow of vehicles. However, synchronizing a set of traffic lights across the road networks of a city is a complex problem that requires solutions in an automatic way. With the aid of a traffic simulator, it is possible to build a computational representation of a set of lights and obtain measures (delay time, travel time, stopped time, average global speed, among others) of traffic-related qualities calculated by the simulator itself. Thus, through the computational representation and a set of quality measures, the problem of traffic light synchronization can be modeled as a Multi-objective Optimization Problem (MOP). Optimization problems that have more than one objective function that must be optimized are called MOPs. Within this class of problems, there has recently been established the Many-Objective Optimization. This area seeks to solve an MOP that has a large number of objective functions, usually problems with more than three functions. In the context of traffic light synchronization, even though the problem is modeled with a large number of objective functions, other studies in the literature seek to optimize only a small subset involving a maximum of two functions. Thus, this study proposes to model and solve the problem of traffic light synchronization as a Many-Objective Optimization Problem (MaOP). In the modeling, the problem was computationally represented and six objective functions were chosen; to solve the problem, many-objective optimization techniques were applied. To model the MaOP, a system was developed and the simulator Simulation of Urban Mobility (SUMO) was used. The purpose of the system is to establish communication between several tasks that are incorporated in modules, so it is possible to communicate between the search algorithm and SUMO. To solve the MaOP, the Nondominated Sorting Genetic Algorithm III (NSGA-III) algorithm and dimensionality reduction techniques were applied, making it possible to model the problem with a reduced number of objectives. In this study, a set of experiments was carried out, aiming to analyze the performance of the NSGA-II and NSGA-III algorithms in different scenarios with many objectives. The results showed that NSGA-II surpassed NSGA-III for the problem in most scenarios, and that dimensionality reduction techniques were effective.