在半导体晶圆制造系统中，具有不确定交货时间窗和不相容产品族的可重入批处理机优化调度是NP-难问题。针对该问题，提出面向产品族的三层滚动混合整数线性规划模型的优化调度算法，其优化目标是提前和误工的时间总和最小。该算法是基于时间序列模型和事件模型的混合分解策略，在可重入批处理机空闲、可用的事件触发下，将整个调度时间轴分解为滚动的时间域；每个时间域对应一个子问题；每个子问题分解为三层：排序、派工和更新；半导体晶圆制造系统实时调度仿真平台用于派工、更新和时域滚动，混合整数线性规划模型用于排序。通过6种典型情形实例验证，结果表明面向产品族的三层滚动混合整数线性规划模型能够在较短计算时间内获得较好优化解。

In the semiconductor wafer fabrication system, for re-entrant batch processing machine with distinct due window and incompatible families, the scheduling problem is known to be NP-complete. Three-phase rolling mixed integer linear programming algorithm is proposed. According to time-sequence-based and event-based combined decomposition strategy, the whole simulation time is decomposed into different rolling horizons. A horizon is corresponding to a sub-problem. Each sub-problem is divided into three phases：Sequencing, dispatching and updating. The real-time scheduling simulation platform and mixed integer linear programming are used to solve dispatching, updating, rolling and sequencing, respectively. 6 representative examples are conducted. The results show that family-oriented three-phase rolling mixed integer linear programming algorithm can obtain better solutions in less computation time.