A robust forecasting technique for inventory and leadtime management
针对随机需求下的库存问题，提出一种基于混合模型的贝叶斯方法，直接估计需求分布高分位点（服务水平），比传统正态模型更鲁棒，适用于基础库存系统和可变提前期的供应管理。
Abstract Inventory stocking problems with stochastic demand typically involve an estimate of the location of some fractile of the demand distribution, where the fractile is usually in the 0.8‐0.99 range. This fractile is termed the “service level” and is the probability that demands will be satisfied from stock on hand. An analogous procedure must be followed to set safety times or time buffers for delivery and supply lead times. The conventional approach of using the Normal model to estimate this location can sometimes be misleading since it primarily uses information about the center of the distribution to predict behavior in the tail of the distribution. A new method based on a mixture model is proposed to estimate the location of the appropriate fractile directly. A formal Bayesian approach is derived and heuristic smoothing methods are developed. Simulation is used to evaluate the methods. The estimate obtained is biased in general but robust in the sense that it works well for a variety of distributions. The performance of this method appears to be superior to the conventional Mean ‐ MAD approach. The method is particularly well suited to application in base‐stock inventory systems and for safety time determination in supply management when leadtimes are variable.