| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| DM_JoãoMoreira_MEI_2022 | 1.44 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Os problemas de planeamento de rotas de veículos, amplamente estudados na literatura, têm uma aplicabilidade real na sociedade. Na recolha de resíduos urbanos pretende-se, por um lado, otimizar a capacidade de transporte dos veículos, e por outro, minimizar a distância por estes percorrida. Deste modo, pretende-se realizar o menor número de rotas possíveis e que garantam a recolha de todos os contentores. Os modelos matemáticos de otimização linear para o problema de encaminhamento de veículos garantem soluções ótimas para a definição de rotas utilizadas na recolha de resíduos, no entanto podem ser computacionalmente difíceis de executar. Na presente investigação foram aplicados métodos de Inteligência Artificial para prever a capacidade ocupada no contentor e, com base nessa previsão, calcular as rotas mais eficientes para efetuar a recolha dos resíduos urbanos, considerando apenas os contentores identificados para recolha. Para isso, foram realizadas comparações entre diversos modelos de Machine Learning, procedendo-se também a ajustes dos respetivos hiper-parâmetros, de forma a obter uma solução eficiente para o cálculo da ocupação nos respetivos contentores. Para o cálculo da rota foram implementadas diferentes abordagens exatas e heurísticas do problema de planeamento de rotas de veículos, de modo a decidir e implementar a melhor abordagem. Como objetivo global desta dissertação, pretende-se apresentar uma boa solução, ou seja, bons tempos computacionais e uma gestão da recolha de resíduos mais eficiente e económica.
Vehicle route planning problems, widely studied in the literature, have real applicability in society. In the collection of urban waste, it is intended, on the one hand, to optimize the transport capacity of vehicles, and on the other, to minimize the distance traveled by them. In this way, it is intended to carry out the fewest possible routes and to guarantee the collection of all containers. The linear optimization mathematical models for the vehicle routing problem guarantee optimal solutions for the definition of routes used in waste collection, however they can be computationally difficult to execute. In the present investigation, Artificial Intelligence methods were applied to predict the occupied capacity in the container and, based on this prediction, calculate the most efficient routes to carry out the collection of urban waste, considering only the containers identified for collection. For this, comparisons were made between different models of Machine Learning, also proceeding with the adjustments of the respective hyper-parameters, in order to obtain an efficient solution for the calculation of the occupancy in the respective containers. For the route calculation, different exact and heuristic approaches to the vehicle route planning problem were implemented, in order to decide and implement the best approach. As a global objective of this dissertation, it is intend to present a good solution, that is, good computational times and a more efficient and economical waste collection management.
Vehicle route planning problems, widely studied in the literature, have real applicability in society. In the collection of urban waste, it is intended, on the one hand, to optimize the transport capacity of vehicles, and on the other, to minimize the distance traveled by them. In this way, it is intended to carry out the fewest possible routes and to guarantee the collection of all containers. The linear optimization mathematical models for the vehicle routing problem guarantee optimal solutions for the definition of routes used in waste collection, however they can be computationally difficult to execute. In the present investigation, Artificial Intelligence methods were applied to predict the occupied capacity in the container and, based on this prediction, calculate the most efficient routes to carry out the collection of urban waste, considering only the containers identified for collection. For this, comparisons were made between different models of Machine Learning, also proceeding with the adjustments of the respective hyper-parameters, in order to obtain an efficient solution for the calculation of the occupancy in the respective containers. For the route calculation, different exact and heuristic approaches to the vehicle route planning problem were implemented, in order to decide and implement the best approach. As a global objective of this dissertation, it is intend to present a good solution, that is, good computational times and a more efficient and economical waste collection management.
Description
Keywords
Gestão de Resíduos Machine Learning Modelos de Previsão Otimização Planeamento de rotas de veículos