| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 7.38 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Diante do contínuo esforço para aprimorar a eficiência dos processos industriais, esta pesquisa foca-se na convergência entre Engenharia e Gestão da Cadeia de Abastecimento e o roteamento. O estudo destaca a urgente necessidade de melhorar o processo de planeamento de rotas, onde a capacidade de prever a rota realmente efetuada pelo motorista se apresenta como um desafio complexo e significativo. A previsão de rotas aplicado à Última Milha é um aspeto crítico da otimização das operações logísticas, onde o foco está na previsão das rotas mais eficientes para a entrega de mercadorias de um centro de distribuição até o destino final, geralmente a localização do cliente. Ao planear rotas com precisão, as empresas podem reduzir os prazos de entrega, diminuir o consumo de combustível e aumentar a satisfação do cliente. Este estudo utiliza abordagens de Aprendizagem Automática para determinar o planeamento de rotas com base nas características específicas de cada uma delas, oferecendo uma solução dinâmica para os desafios do roteamento. Ao mesmo tempo, esta pesquisa incentiva a colaboração entre a Inteligência Artificial e o setor industrial, aumentando a eficiência em aplicações fundamentais. Portanto, o objetivo principal deste projeto é desenvolver, um modelo de previsão de rotas capaz de tentar atingir a rota realmente efetuada pelos motoristas. Para alcançar isso foi necessário realizar previamente uma análise minuciosa aos dados fornecidos, de modo a encontrar padrões relevantes e assim extrair informação valiosa para o negócio.
Esta dissertação utiliza a metodologia Cross Industry Standard Process for Data Mining (CRISP-DM)
como estrutura guia. Ao investigar a sua natureza adaptativa, os algoritmos de Aprendizagem Automática mostram um potencial significativo para aumentar tanto a eficiência quanto a precisão no processo de previsão de rotas. Este estudo dado ser baseado na frequência dos motoristas, iniciou-se com a criação de um baseline
servindo como um ponto de referência com o qual se vai comparar os resultados futuros, sendo depois mais à frente importante para avaliar o desempenho dos modelos. De seguida, realizou-se uma análise minuciosa de duas abordagens diferentes, com ênfase na avaliação da consistência e precisão das previsões feitas, sendo que daí resultou a comparação entre três algoritmos o baseline, o OSquare e o Routyn. Uma das abordagens pautou-se na modelação baseada em histórico de frequências. A outra abordagem baseou-se na adaptação do algoritmo OSquare, sendo que daqui se destacou o XGBoost como modelo utilizado, visto que obteve os melhores resultados. Além disso, dada a pesquisa acerca da literatura o OSquare era o que melhor se enquadrava para o problema em questão. Daqui resultou uma comparação entre vários algoritmos o baseline, o OSquare e o Routyn, em que este último é o utilizado pela empresa para obtenção dos percursos a
efetuar nas rotas. Utilizando as métricas de Coeficiente de Correlação de Kendall, Edit Distance e Accuracy chegou-se à conclusão, que apesar da amostra não ser significativa, o OSquare foi o que melhor conseguiu aproximar as rotas efetuadas pelo motorista, sendo algo de realçar. Esta dissertação, além de ampliar o conhecimento na área de pesquisa, evidencia claramente o potencial das abordagens de Aprendizagem Automática na previsão de rotas. Dessa forma, faz uma contribuição significativa neste campo, ressaltando a importância da análise de dados e da modelagem como métodos capazes para enfrentar os desafios contínuos desta indústria em constante mudança.
Given the continuous effort to improve the efficiency of industrial processes, this research focuses on the convergence between Supply Chain Engineering and Management and routing. The study highlights the urgent need to improve the route planning process, where the ability to predict the route actually taken by the driver presents a complex and significant challenge. Route forecasting applied to Last Mile is a critical aspect of optimizing logistics operations, where the focus is on predicting the most efficient routes for delivering goods from a distribution center to the final destination, usually the customer's location. By accurately planning routes, companies can reduce delivery times, decrease fuel consumption and increase customer satisfaction. This study uses Machine Learning approaches to determine route planning based on the specific characteristics of each route, offering a dynamic solution to routing challenges. At the same time, this research encourages collaboration between Artificial Intelligence and the industrial sector, increasing efficiency in fundamental applications. Therefore, the main objective of this project is to develop a route prediction model capable of trying to reach a route actually taken by drivers. To achieve this, it was necessary to previously carry out a thorough analysis of the data provided, in order to find relevant patterns and thus extract valuable information for the business. This dissertation uses the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology as a guiding structure. By investigating their adaptive nature, Machine Learning algorithms show significant potential to increase both efficiency and accuracy in the route prediction process. T This study, given that it is based on the frequency of drivers, began with the creation of a baseline, serving as a reference point with which future results will be compared, and later on being important for evaluating the performance of the models. Next, a thorough analysis of two different approaches was carried out, with an emphasis on evaluating the consistency and accuracy of the predictions made, resulting in a comparison between three algorithms: baseline, OSquare and Routyn. One of the approaches was based on modeling based on frequency history. The other approach was based on the adaptation of the OSquare algorithm, from which XGBoost stood out as the model used, as it obtained the best results. Furthermore, given the literature research, OSquare was the best fit for the problem in question. This resulted in a comparison between several algorithms, baseline, OSquare and Routyn, the latter being used by the company to obtain the routes to be taken on the routes. Using the Kendall Correlation Coefficient, Edit Distance and Accuracy metrics, it was concluded that although the sample was not significant, OSquare was the one that best approximated the routes taken by the driver, which is something worth highlighting. This dissertation, in addition to expanding knowledge in the research area, clearly highlights the potential of Machine Learning approaches in route prediction. In this way, it makes a significant contribution to this field, highlighting the importance of data analysis and modeling as capable methods to face the ongoing challenges of this ever-changing industry.
Given the continuous effort to improve the efficiency of industrial processes, this research focuses on the convergence between Supply Chain Engineering and Management and routing. The study highlights the urgent need to improve the route planning process, where the ability to predict the route actually taken by the driver presents a complex and significant challenge. Route forecasting applied to Last Mile is a critical aspect of optimizing logistics operations, where the focus is on predicting the most efficient routes for delivering goods from a distribution center to the final destination, usually the customer's location. By accurately planning routes, companies can reduce delivery times, decrease fuel consumption and increase customer satisfaction. This study uses Machine Learning approaches to determine route planning based on the specific characteristics of each route, offering a dynamic solution to routing challenges. At the same time, this research encourages collaboration between Artificial Intelligence and the industrial sector, increasing efficiency in fundamental applications. Therefore, the main objective of this project is to develop a route prediction model capable of trying to reach a route actually taken by drivers. To achieve this, it was necessary to previously carry out a thorough analysis of the data provided, in order to find relevant patterns and thus extract valuable information for the business. This dissertation uses the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology as a guiding structure. By investigating their adaptive nature, Machine Learning algorithms show significant potential to increase both efficiency and accuracy in the route prediction process. T This study, given that it is based on the frequency of drivers, began with the creation of a baseline, serving as a reference point with which future results will be compared, and later on being important for evaluating the performance of the models. Next, a thorough analysis of two different approaches was carried out, with an emphasis on evaluating the consistency and accuracy of the predictions made, resulting in a comparison between three algorithms: baseline, OSquare and Routyn. One of the approaches was based on modeling based on frequency history. The other approach was based on the adaptation of the OSquare algorithm, from which XGBoost stood out as the model used, as it obtained the best results. Furthermore, given the literature research, OSquare was the best fit for the problem in question. This resulted in a comparison between several algorithms, baseline, OSquare and Routyn, the latter being used by the company to obtain the routes to be taken on the routes. Using the Kendall Correlation Coefficient, Edit Distance and Accuracy metrics, it was concluded that although the sample was not significant, OSquare was the one that best approximated the routes taken by the driver, which is something worth highlighting. This dissertation, in addition to expanding knowledge in the research area, clearly highlights the potential of Machine Learning approaches in route prediction. In this way, it makes a significant contribution to this field, highlighting the importance of data analysis and modeling as capable methods to face the ongoing challenges of this ever-changing industry.
Description
Keywords
Route prediction Machine learning Last mile delivery OSquare Frequency matrix Aprendizagem automática Última milha Matriz de frequências Previsão de rotas