Per-instance algorithm configuration for production planning in a reconfigurable assembly system

Reconfigurable assembly systems (RAS) are designed to operate in highly volatile market environments, which establishes the need for constant redefinitions of the production planning activities. These planning activities are usually modeled as complex mathematical optimization problems, and as such, increased responsiveness of the system will be directly linked to the prompt and efficient solution of these problems. In this work, we investigate the use of Per-instance algorithm configuration (PIAC) methods to select, at runtime, the best configuration of an algorithm designed to solve these optimization problems. We compare and evaluate the performance of Hydra, a state-of-the-art PIAC method designed for heterogeneous instance spaces, and a simpler Case Base Reasoning (CBR) approach to PIAC. Our experiments suggest that CBR methods for PIAC could be more suitable to highly homogeneous instance spaces, such as those observed in a RAS context, with reductions of up to 17% over the default algorithm configuration settings.