Multi-Objective Genetic Programming based on Decomposition on Evolving Scheduling Heuristics for Dynamic Scheduling

Abstract

Dynamic flexible job shop scheduling (DFJSS) is an important combinatorial optimisation problem that requires handling machine assignment and operation sequencing simultaneously in dynamic environments. Genetic programming (GP) has achieved great success to evolve scheduling heuristics for DFJSS. In manufacturing, multi-objective DFJSS (MO-DFJSS) is more common and challenging due to conflicting objectives. Existing Pareto dominance-based multi-objective GP methods show their limitations of not providing good spreadability and consistency in heuristic behaviour. Multi-objective evolutionary algorithm based on decomposition (MOEA/D) has the potential to provide good spreadability and consistency due to the mechanisms of weights-based subproblems decomposition and neighbours-based evolution. However, it is non-trivial to apply MOEA/D to MO-DFJSS since we need to search in heuristic space. To address these challenges, we propose a multi-objective GP approach based on decomposition (MOGP/D) that incorporates the advantages of MOEA/D and GP to learn scheduling heuristics for MO-DFJSS. A mapping strategy is designed to find the fittest individual for each subproblem. Extensive experiments show that MOGP/D obtains competitive performance with the state-of-the-art methods for MO-DFJSS, and good spreadability and consistency in heuristic behaviour.