基于支持向量机的半导体生产线动态调度方法

Dynamic scheduling approach based on SVM for semiconductor production line

为了快速合理地选择调度策略,研究了一种半导体生产线动态调度策略选择方法。该方法以历史数据为基础,选取支持向量机为数据挖掘工具,采用二进制粒子群优化算法对生产属性(特征)子集进行寻优,获得基于支持向量机的动态调度策略分类模型。对于任意给定的生产状态,通过该模型,能实时地获取当前生产状态下近似最优的调度策略。在调度策略评价中,选用了基于功效函数与熵权法的多目标评价方法,以扩展该方法的应用范围。在某实际硅片生产线上验证了所提动态调度方法的有效性。

A dynamic dispatching strategy selection approach for semiconductor production line was researched for quick and reasonable choice of scheduling strategies. Based on historical data, the Support Vector Machine （SVM） was selected as a data mining tool to obtain SVM-based dynamic scheduling strategy classification model with for production line by optimizing production attributes subset with Binary Particle Swarm Optimization（BPSO） algo- rithm. Under any given production status, an approximate optimal scheduling strategy could be acquired through this model in real-time. In the evaluation of scheduling strategies, a multi-object evaluation method based on power functions and entropy weight method was employed to extend the approach＇s application range. The effectiveness and feasibility of proposed dynamic scheduling approach was tested in an actual semiconductor production line.