Speeding up a Rollout algorithm for complex parallel machine scheduling
针对Rollout算法计算时间长的缺点，提出动态启发式剪枝和候选缩减策略来加速，并在复杂并行机器调度问题上验证了有效性。
Rollout methodology is a constructive metaheuristic algorithm and its main characteristics are its modularity, the adaptability to different objectives and constraints and the easiness of implementation. Multi-heuristic Rollout extends the Rollout by incorporating several constructive heuristics in the Rollout framework and it is able to easily incorporate human experience inside its research patterns to fulfil complex requirements dictated by the application at hand. However, a drawback for both Rollout and multi-heuristic Rollout is often represented by the required computation time. This paper proposes some alternatives of the full multi-heuristic Rollout algorithm aimed at improving the efficiency by reducing the computational effort while preserving the effectiveness. Namely, we propose dynamic heuristics pruning and candidates reduction strategies. As illustrative case studies, we analyse complex deterministic identical parallel machine scheduling problems showing how Rollout procedures can be used to tackle several additional constraints arising in real contexts. More specifically, we considered both standard (batch production, family set-ups, release, due dates, etc.) and non-standard (machine unavailabilities, maximum campaign size) scheduling constraints. An extensive campaign of computational experiments shows the behaviour of the multi-heuristic Rollout approach and the effectiveness of the different proposed speed-up methods.