Inclusión de la robustez en la optimización multi-objetivo para problemas de secuenciación de tareas tipo Job Shop
Archivos
Fecha
2014-06-25
Autores
Rodríguez Bazán, Erick David
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
Durante las últimas décadas, los algoritmos multi-objetivo han captado considerable atención. El problema de secuenciación tipo Job Shop, uno de los principales problemas de optimización, también ha capturado la atención. Sin embargo, la mayoría de los algoritmos multi-objetivos que se han desarrollado para este problema están basados en enfoque no profesionales.es decir, la mayoría de ellos combinan sus objetivos y resuelven el problema multi-objetivo a través de enfoques basados en un solo objetivo, exceptuando a un pequeño grupo de investigadores que utilizan algoritmos basados en la Frontera de Pareto. En esta tesis se consideran dos objetivos, makespan y robustez. La robustez es dada por el valor esperado de la diferencia relativa entre el makespan determinístico y el real. Se implementaron dos medidas para la robustez. La primera medida está basada en la probabilidad de que una maquina se rompa, mientras la segunda se basa en intervalos críticos de acuerdo a un umbral. Se implementó un algoritmo basado en la Frontera de Pareto usando Aprendizaje Reforzado. El algoritmo propuesto se aplicó a un conjunto de instancias para optimizar los objetivos tratados y finalmente los resultados experimentales muestran que la secuenciación robusta asociada al makespan óptimo no siempre es la más robusta, en algunas ocasiones es mejor sacrificar optimalidad y ganar en robustez.
During the last decades, developing multi-objective algorithms for optimization problems has found considerable attention. Job Shop scheduling problem, as one of the main important scheduling optimization problem, has found this attention too. However, most of the multi-objective algorithms that have been developed for this problem use nonprofessional approaches. In another words, most of them combine their objectives and then solve multi-objective problem through single objective approaches, except some scarce researcher that uses Pareto-based algorithms. In this paper, two objectives – makespan and robustness – are simultaneously considered. Robustness is indicated by the expected value of the relative difference between the deterministic and actual makespan. Two measures for robustness are developed; the first measure is based on the probability of machine breakdowns, while the second measure is based in critic intervals according to a threshold. To address this problem, a multi-objective algorithm using reinforcement learning and based in the Pareto Front is presented. The proposed algorithm is applied to benchmark
During the last decades, developing multi-objective algorithms for optimization problems has found considerable attention. Job Shop scheduling problem, as one of the main important scheduling optimization problem, has found this attention too. However, most of the multi-objective algorithms that have been developed for this problem use nonprofessional approaches. In another words, most of them combine their objectives and then solve multi-objective problem through single objective approaches, except some scarce researcher that uses Pareto-based algorithms. In this paper, two objectives – makespan and robustness – are simultaneously considered. Robustness is indicated by the expected value of the relative difference between the deterministic and actual makespan. Two measures for robustness are developed; the first measure is based on the probability of machine breakdowns, while the second measure is based in critic intervals according to a threshold. To address this problem, a multi-objective algorithm using reinforcement learning and based in the Pareto Front is presented. The proposed algorithm is applied to benchmark
Descripción
Palabras clave
Multi-Objetivo, Secuenciación, Job Shop, Aprendizaje Reforzado, Frontera de Pareto, Multiple Objectives, Scheduling, Pareto Front, Reinforcement Learning