Solución al problema de secuenciación en máquinas paralelas utilizando aprendizaje reforzado
Archivos
Fecha
2010-07-10
Autores
Suárez Ferreira, Juliett Maybetsy
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
Los problemas de Optimización Combinatoria ocupan diversos campos como la economía, el comercio, la ingeniería, la industria o la medicina. Los problemas de secuenciación, como ejemplo de estos, consisten en la localización o asignación de recursos en el tiempo a un conjunto de tareas o actividades; dentro de ellos, aparece el problema de secuenciación en máquinas paralelas, caracterizado como la actividad de asignar un número de trabajos a un conjunto de máquinas con el objetivo de minimizar el costo total de procesamiento.
Recientemente, el aprendizaje reforzado ha sido utilizado para resolver problemas de secuenciación reportando buenos resultados, pero su uso no se ha extendido a problemas de secuenciación complejos donde existen distintos tipos de recursos disponibles. Considerando lo anterior, en este trabajo se presenta una alternativa de solución al problema de secuenciación en máquinas paralelas (JSSP-PM) utilizando el algoritmo Q-Learning del Aprendizaje Reforzado. En esta propuesta se realizan adaptaciones al algoritmo original para ajustarlo a las características del problema y se introducen dos enfoques para dar solución al mismo.
Finalmente, el estudio experimental se realiza utilizando instancias del problema que se encuentran disponibles en la librería de investigación de operaciones. Los resultados obtenidos por las variantes propuestas son comparados entre ellos y además con los reportados por otros enfoques.
Combinatorial Optimization problems are present in different fields: economy, commerce, engineering, industry or medicine. Scheduling problems, for example, consist in the allocation of resources to jobs or activities; one of these scheduling problems is the Job Shop Scheduling Problem with Parallel Machines (JSSP-PM), which can be characterized as the activity of assigning a number of jobs to a set of machines such that the total processing time is minimized. Recently, Reinforcement Learning has been applied in the solution of scheduling problems reporting very good results, but it has not been applied to more complex scheduling problems, where different types of resources are available. Considering this, the present work introduces an alternative to solve the Job Shop Scheduling Problem with Parallel Machines using the Q-Learning algorithm from the Reinforcement Learning field. In this proposal, some adaptations to the original algorithm are made in order to adapt it to the characteristics of the problem and two approaches are introduced in order to solve it. Finally, the experimental study is developed using some instances of the problem available in the Operational Research Library. The results obtained by the alternatives proposed are compared between them and also with the results reported by some other approaches.
Combinatorial Optimization problems are present in different fields: economy, commerce, engineering, industry or medicine. Scheduling problems, for example, consist in the allocation of resources to jobs or activities; one of these scheduling problems is the Job Shop Scheduling Problem with Parallel Machines (JSSP-PM), which can be characterized as the activity of assigning a number of jobs to a set of machines such that the total processing time is minimized. Recently, Reinforcement Learning has been applied in the solution of scheduling problems reporting very good results, but it has not been applied to more complex scheduling problems, where different types of resources are available. Considering this, the present work introduces an alternative to solve the Job Shop Scheduling Problem with Parallel Machines using the Q-Learning algorithm from the Reinforcement Learning field. In this proposal, some adaptations to the original algorithm are made in order to adapt it to the characteristics of the problem and two approaches are introduced in order to solve it. Finally, the experimental study is developed using some instances of the problem available in the Operational Research Library. The results obtained by the alternatives proposed are compared between them and also with the results reported by some other approaches.
Descripción
Palabras clave
Problema de Secuenciación, Máquinas Paralelas, Aprendizaje Reforzado, Estudio Experimental