Algoritmos para problemas de secuenciación de tareas en ambientes online
Archivos
Fecha
2015-07-26
Autores
Coto Palacio, Jessica
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
El problema de secuenciación de tareas es usualmente definido como “el problema de acomodar los recursos en el tiempo para realizar un conjunto de tareas". Algunos de esos problemas ocurren en ambientes online, debido a la necesidad de secuenciar con información incompleta, pues en ciertos puntos las decisiones deben ser tomadas sin tener conocimiento sobre eventos futuros.
El uso de agentes y de aprendizaje reforzado combinado con reglas de despacho para dar solución a este tipo de problemas resulta un enfoque poco explorado, de ahí la idea de utilizar dichas técnicas en la solución del mismo, ya que este tipo de aprendizaje es uno de los que está siendo actualmente aplicado en la solución de problemas de secuenciación de forma offline. El aprendizaje de los agentes en cada punto de decisión estará determinado por un algoritmo de aprendizaje, y para esto se implementaron dos alternativas, el Q-Learning y el Learning Automata.
El enfoque de solución a problemas de secuenciación de tareas en ambientes online propuesto en esta tesis se basa en un caso de estudio existente en la literatura, y a partir del mismo se definen otros escenarios de producción. Los resultados obtenidos son validados utilizando pruebas estadísticas, a partir de 315 instancias, divididas por escenario y combinación posible de cada algoritmo con varios parámetros reportados en la literatura.
Scheduling problems are usually defined as “the problem of allocating resources in time in order to perform a set of tasks”. Some of these problems occur in online environments, due to the need of scheduling with incomplete information, because in certain states decision have to be made without knowledge about future events. The use of agents and reinforcement learning combined with dispatching rules in order to solve this kind of problems is a poorly explored approach, that is why we decided to use these techniques for its solution, because this type of learning is being widely used in the solution of offline scheduling problems. The learning process of the agents in each decision point will be determined by a learning algorithm, and for this we have two alternatives, Q-Learning and Learning Automata. The approach to solve online scheduling problems proposed in this thesis is based on a case study from the literature, and starting from this basic idea we define other production scenarios. The results are validated using statistical tests, for this we use 315 instances, divided by scenario and also by possible combination of each algorithm with different parameters reported in the literature.
Scheduling problems are usually defined as “the problem of allocating resources in time in order to perform a set of tasks”. Some of these problems occur in online environments, due to the need of scheduling with incomplete information, because in certain states decision have to be made without knowledge about future events. The use of agents and reinforcement learning combined with dispatching rules in order to solve this kind of problems is a poorly explored approach, that is why we decided to use these techniques for its solution, because this type of learning is being widely used in the solution of offline scheduling problems. The learning process of the agents in each decision point will be determined by a learning algorithm, and for this we have two alternatives, Q-Learning and Learning Automata. The approach to solve online scheduling problems proposed in this thesis is based on a case study from the literature, and starting from this basic idea we define other production scenarios. The results are validated using statistical tests, for this we use 315 instances, divided by scenario and also by possible combination of each algorithm with different parameters reported in the literature.
Descripción
Palabras clave
Algoritmos, Problemas de Secuenciación, Conjunto de Tareas, Ambientes Online, Agentes, Aprendizaje Reforzado, Q-Learning, Learning Automata, Caso de Estudio, Inteligencia Artificial