Paper Title
Machine Learning in Hybrid Flow Shop Scheduling with Unrelated Machines

Hybrid flow shop (HFS) problems are often encountered in real world production systems. Despite their practical relevance, very few generic methods exist to solve HFS problems. Instead, many approaches either focus on two-stage problems or the application of simple dispatching rules. Moreover, the majority of authors assume identical parallel machines, which reduces the complexity of machine assignment to a large extent. If a real world case is studied, solution methods are often customized and not adaptable to other settings. Most common decomposition approaches are critical pathrelated or only focus on possible permutations of job indices while utilizing simple machine assignment rules. [1] Lately, machine learning (ML) has been applied to different scheduling problems. A common application is the selection of best dispatching rules based on the state of systems parameters. We propose an alternative ML-based approach to makespan or flow time minimization that is suitable for different configurations regarding the number of stages and number of unrelated machines per stage. To speed up the scheduling process, we apply ML in one step of our solution method: We train Neural Networks (NN) and Support Vector Machines (SVM) with optimal machine assignments and makespan or flowtime values for fixed batch sizes and randomly generated processing time matrices. Afterwards, we use the trained NN and SVM to predict optimal makespan or flowtime values and machine assignments for all partial job sequences (batches) based on a given processing time matrix. Only sequences with close-to-optimal makespan values are evaluated further to determine a final machine assignment and overall sequence. Keywords - Hybrid Flow Shop, Unrelated Machines, Deterministic Scheduling, Divide Et Impera, Machine Learning.