Online optimization with gradual look-ahead
Online optimization with gradual look-ahead
复制标题
逐步前瞻的在线优化
作者：
Fabian Dunke;S. Nickel
我们扩展设置的在线优化与前瞻在线优化逐步前瞻。虽然到目前为止所考虑的前瞻是指对未来数据的确定性展望，但渐进前瞻仅允许对未来数据的不确定性展望，其随着输入元素的发布时间的接近而变得越来越精确。在讨论了相关概念之后，我们正式介绍了具有渐进前瞻的在线优化问题。由于一个单一的输入元素的前瞻性信息的过程中被绑定到一个相应的不确定性集的轨迹，我们研究如何不同的预测方法和不同的算法方法来处理逐步前瞻性可以被实例化，并相互比较的优化输出。我们通过数值实验验证了所引入的概念，在逐步前瞻下的批量和车辆路径的两个应用。
We extend the setting of online optimization with look-ahead to online optimization with gradual look-ahead. While look-ahead as considered so far refers to a deterministic outlook on future data, gradual look-ahead only allows for an uncertain outlook on future data which becomes more and more precise as an input element’s release time is approached. After a discussion of related concepts, we formally introduce the class of online optimization problems with gradual look-ahead. Since the course of look-ahead information of a single input element is tied to a corresponding uncertainty set trajectory, we examine how different forecasting methods and different algorithmic approaches for dealing with gradual look-ahead can be instantiated and compared to each other with respect to the optimization output. We exemplify the introduced concepts by numerical experiments for the two applications of lot sizing and vehicle routing under gradual look-ahead.