Publicação
Optimization of public transport networks: Reinforcement learning for smart mobility
| Resumo: | Designing an efficient public transport network is a complex problem (NP-hard) that involves, among other factors, selecting stops, determining the most optimal routes, and defining schedules and frequencies, all while considering multiple conflicting factors. Given this complexity, the pursuit of optimal solutions is often set aside in favor of heuristic methods (general decision rules) and expert knowledge, which allow for identifying satisfactory solutions. This dissertation focuses on optimizing Lisbon’s Carris public transport network by exploring the application of reinforcement learning (RL) mechanisms to address part of this problem: finding more optimal routes between several stops served by a variable number of lines – a vehicle routing problem. Two models were trained using the Multi-task Vehicle Routing Solver with Mixture-of-Experts algorithm. The results were compared with the Carris network and the results given by the Clarke and Wright Savings algorithm. RL shows potential in learning good heuristics and finding better solutions than the current ones, as the model minimized the straight-line distance of the shortest segment of the network. However, the complexity of urban mobility remains a challenge, requiring simplifications to model this problem effectively. Despite limitations such as low computational resources and the static nature of the data, this analysis demonstrates that by integrating traffic information and developing more comprehensive algorithms, RL can improve the efficiency of these networks and create solutions that dynamically adjust to different daily constraints. |
|---|---|
| Autores principais: | Mota, António Luís Barros |
| Assunto: | Reinforcement learning VRP Lisbon Carris Smart mobility Rede de transporte -- Transport network Transporte público -- Public transportation Aprendizagem reforçada Lisboa Mobilidade inteligente |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | ISCTE |
| Idioma: | inglês |
| Origem: | Repositório ISCTE |
| Resumo: | Designing an efficient public transport network is a complex problem (NP-hard) that involves, among other factors, selecting stops, determining the most optimal routes, and defining schedules and frequencies, all while considering multiple conflicting factors. Given this complexity, the pursuit of optimal solutions is often set aside in favor of heuristic methods (general decision rules) and expert knowledge, which allow for identifying satisfactory solutions. This dissertation focuses on optimizing Lisbon’s Carris public transport network by exploring the application of reinforcement learning (RL) mechanisms to address part of this problem: finding more optimal routes between several stops served by a variable number of lines – a vehicle routing problem. Two models were trained using the Multi-task Vehicle Routing Solver with Mixture-of-Experts algorithm. The results were compared with the Carris network and the results given by the Clarke and Wright Savings algorithm. RL shows potential in learning good heuristics and finding better solutions than the current ones, as the model minimized the straight-line distance of the shortest segment of the network. However, the complexity of urban mobility remains a challenge, requiring simplifications to model this problem effectively. Despite limitations such as low computational resources and the static nature of the data, this analysis demonstrates that by integrating traffic information and developing more comprehensive algorithms, RL can improve the efficiency of these networks and create solutions that dynamically adjust to different daily constraints. |
|---|