Essays on urban bus transport optimization

Abstract

Nesta tese, nós apresentamos uma compilação de três artigos de otimização aplicados no contexto de transporte urbano de ônibus. O principal objetivo foi estudar e implementar heurísticas com base em Pesquisa Operacional para otimizar problemas de (re)escalonamento de veículos off-line e on-line considerando várias garagens e frota heterogênea. No primeiro artigo, foi proposta uma abordagem heurística para o problema de escalonamento de veículos múltiplas garagens. Acreditamos que as principais contribuições são o método de geração de colunas para grandes instâncias e as técnicas de redução do espaço de estados para acelerar as soluções. No segundo artigo, adicionamos complexidade ao considerar a frota heterogênea, denotada como multiple depot vehicle type scheduling problem (MDVTSP). Embora a importância e a aplicabilidade do MDVTSP, formulações matemáticas e métodos de solução para isso ainda sejam relativamente inexplorados. A principal contribuição desse trabalho foi o método de geração de colunas para o problema com frota heterogênea, já que nenhuma outra proposta na literatura foi identificada no momento pelos autores. Na terceira parte desta tese, no entanto, nos concentramos no reescalonamento em tempo real para o caso de quebras definitivas de veículos. A principal contribuição é a abordagem eficiente do reescalonamento sob uma quebra. A abordagem com redução de espaço de estados, solução inicial e método de geração de colunas possibilitou uma ação realmente em tempo real. Em menos de cinco minutos, reescalonando todas as viagens restantes.

In this dissetation we presented a three articles compilation in urban bus transportation optimization. The main objective was to study and implement heuristic solutions method based on Operations Research to optimizing offline and online vehicle (re)scheduling problems considering multiple depots and heterogeneous fleet. In the first paper, a fast heuristic approach to deal with the multiple depot vehicle scheduling problem was proposed. We think the main contributions are the column generation framework for large instances and the state-space reduction techniques for accelerating the solutions. In the second paper, we added complexity when considering the heterogeneous fleet, denoted as "the multiple-depot vehicle-type scheduling problem" (MDVTSP). Although the MDVTSP importance and applicability, mathematical formulations and solution methods for it are still relatively unexplored. We think the main contribution is the column generation framework for instances with heterogeneous fleet since no other proposal in the literature has been identified at moment by the authors. In the third part of this dissertation, however, we focused on the real-time schedule recovery for the case of serious vehicle failures. Such vehicle breakdowns require that the remaining passengers from the disabled vehicle, and those expected to become part of the trip, to be picked up. In addition, since the disabled vehicle may have future trips assigned to it, the given schedule may be deteriorated to the extent where the fleet plan may need to be adjusted in real-time depending on the current state of what is certainly a dynamic system. Usually, without the help of a rescheduling algorithm, the dispatcher either cancels the trips that are initially scheduled to be implemented by the disabled vehicle (when there are upcoming future trips planned that could soon serve the expected demand for the canceled trips), or simply dispatches an available vehicle from a depot. In both cases, there may be considerable delays introduced. This manual approach may result in a poor solution. The implementation of new technologies (e.g., automatic vehicle locators, the global positioning system, geographical information systems, and wireless communication) in public transit systems makes it possible to implement real-time vehicle rescheduling algorithms at low cost. The main contribution is the efficient approach to rescheduling under a disruption. The approach with integrated state-space reduction, initial solution, and column generation framework enable a really real-time action. In less than five minutes rescheduling all trips remaining.