Production scheduling in the textile industry: a practical approach using evolutionary algorithms with domain-dependent information

Abstract

The term refers to a class of iterative search methods which mimic evolutionary processes in nature. They include genetic algorithms, evolutionary strategies and evolutionary programming. Principles of natural selection, crossover and mutation are implemented on a population of alternative, valid solutions in order to find good solutions on problems with very large search spaces. Up until now, there have been relatively few implementations of evolutionary algorithms on large scheduling problems in industry. One reason may be the large amount of computer processing time required to find acceptable solutions. This paper details the implementation of an evolutionary algorithm which makes use of domain-dependent information to generate production schedules for a manufacturer in the textile industry. In order to efficiently process large complex data sets, a nonbinary chromosome representation which utilizes hierarchical, dynamic data structures is presented. Special, domain-dependent crossover and mutation operators are proposed to ensure that only valid solutions are generated. In addition, aspects of tabu search are used as meta-heuristics to improve the performance of the evolutionary algorithm. These methods are integrated into a complete production planning and control system for a textile manufacturer.