Abstract: The due-dates of mould projects were hard to predict caused by various uncertain factors.Concerning with this problem，several major random factors in mould manufacturing were analyzed and the multi-mode concept was proposed.Moreover，a stochastic prediction model was established by combining dynamic workshop load control theory.Based on this model，the evolution of ongoing projects was conducted by integrating Multi-mode Resource Constrain Projects Schedule Problem (MRCPSP) priority rules and the mathematical distribution of due-date was obtained.During this process，dynamic programming was used to obtain optimal strategies which minimized penalty of project tardiness.To reduce computational complexity，the state space was confined by Markov rough evolution and eliminated to low occurring probabilities.A project management module was developed and applied in the evolution of a project group，the distribution of due-date and optimal strategies were obtained.The result showed that the prediction model had good practicability.

Key words: multi-mode, Markov decision process, dynamic programming, due-date forecast, priority rules, mould manufacturing

摘要： 针对模具制造项目由不确定性因素导致的交货期难以预测的问题，分析了模具生产过程中的几种主要随机因素，并提出多模式的概念。结合车间动态负荷控制的理念，建立了在制项目的交货期随机预测模型。在该模型的基础上，通过融入多模式资源受限项目调度优先规则，对在制项目进行演化，得到项目交货期数学分布，同时用动态规划求解最优策略集，以使项目群拖期惩罚最小。在演化计算过程中，为有效减少计算负荷，采用马氏链粗演化及屏蔽低概率状态的方法缩小系统状态空间。开发了一个项目管理模块，成功地对一个项目群进行演化，获得了项目交货期概率分布及最佳策略，结果表明了该预测模型的实用性。

关键词: 多模式, Markov决策过程, 动态规划, 交货期预测, 优先规则, 模具制造