ALGORITMOS PARA PROBLEMAS DE SECUENCIACIÓN DE TAREAS EN AMBIENTES ONLINE

Abstract

Scheduling problems are present in many processes that occur in the manufacturing industry, where it is necessary to perform a set of
operations at certain periods of time and it also needs to allocate limited resources to perform these tasks. Some of these problems occur
in online environments, because there is no prior knowledge of the arrival of the jobs or the time it would take for each job to be processed
in each of the machines. In this paper we study and propose a solution to online scheduling problems based on an existing case study from the literature, using two Reinforcement Learning algorithms. Besides we extend the study case to two more complex scenarios. The results obtained show the superiority of Q-Learning algorithm over Learning Automata algorithm due to the flexibility of Q-Learning’s
parameters. These results were validated using statistical tests where an algorithm is better than other if its difference between the generated and the processed jobs is less. Friedman test applied among all variants to find significant differences and also we applied Wilcoxon test to determinate the best algorithm by scenario.