An island-based optimization framework for process planning and scheduling using migrations with empirical linkage

The integrated process planning and scheduling problem is widely present in the industrial field, which is recognized as an NP-hard problem and is often tackled by optimization approaches such as Genetic Algorithms (GA). To provide sufficient computational resources that are often not available in real-world scenarios, a cloud-based GA framework with the island model (IM) is applied, in which the population is divided into several subpopulations (i.e., islands) that are evolved independently. However, existing IMs either do not support inter-island communication or require synchronization of all islands with a non-trivial cost, which significantly undermines the optimization quality. This paper presents a novel asynchronous island-based optimization framework for integrated process planning and scheduling. The asynchronous IM eliminates the synchronization cost due to inter-island communication via migrations and supports a wide range of migration strategies to improve optimization efficiency. Two types of migration operators are constructed. The first, classical one, migrates a complete individual (i.e., individual-based migration). In the second approach, a finer-grained migration is constructed that migrates only the highly-dependent groups of genes by discovering the dependencies between genes in an individual using empirical linkage learning. This approach leads to slower convergence but is less likely to get trapped in local optima, which is significantly more effective than IMs using any of the considered individual-based migration operators. Evaluation based on a real-world manufacturing process planning and scheduling problem shows that the proposed method significantly outperforms existing methods.