Otimização com aprendizado autônomo para programação da produção com sequências de instâncias heterogêneas
Abstract
Many scheduling problems, but also in Advanced Planning and Scheduling (APS), are
NP-Hard. This project addresses a scheduling problem in a single machine environment
with sequence dependent setup times. It is possible to partially outsource the demand
restricting that there are no delays for delivering of the orders, aiming to minimize the
cost for outsourcing. Evolutionary Algorithms (EA) represent a fast solution strategy
for NP-Hard problems. However, researchers in the field of evolutionary computation
say that EAs depend significantly on the configuration of their parameters. This work
investigates in the scheduling and APS literature, how the authors who develop EAs
determines the parameterization and evaluation of their algorithms, also presenting
which strategies are suggested as state of the art for automatic tuning of EAs. This
thesis states an innovative strategy for automated configuration of EAs, including
a new paradigm for optimization in streams of heterogeneous instances called ALO
(Autonomous Learning Optimization). This new paradigm aims to solve integratelly
an optimization problem and parameterization of the configurable algorithm with
an autonomous decision process for detecting heterogeneities within the sequence of
instances.