Machine scheduling instance generation by reverse engineering from instance space analysis

The availability of a sufficiently large number of meaningful instances for a scheduling problem is of utmost importance for the evaluation of solution methods for the problem. This study introduces a novel method for machine scheduling instance generation, termed Reverse Instance Generation (RIG), leveraging Instance Space Analysis. This method aims to create diverse, feasible, and realistic instances by reverse engineering from the instance space. Unlike existing approaches that rely on iterative search methods, RIG utilizes a constructive approach, combining dimensionality reduction techniques and controlled instance generation. The approach addresses the challenges of instance diversity and reasonableness, ensuring unbiased and reproducible outcomes. The effectiveness of RIG is demonstrated on three different machine scheduling problems: the single-machine weighted tardiness problem, the job shop scheduling problem, and a complex serial batch scheduling problem. The results highlight the method's ability to cover gaps in the instance space while maintaining practicality and efficiency, paving the way for improved benchmarking and algorithm development.