Abstract: In the context of intelligent manufacturing,information technologies such as the Internet of things have accumulated a large amount of data for the manufacturing system.Meanwhile,advanced methods such as artificial intelligence provide effective means for data analysis and real-time control of shop floor.Therefore,a production-data-based dynamic scheduling method was proposed to minimize the makespan for the hybrid flow shop scheduling problem with unrelated parallel machines.The production features and scheduling rules were extracted to complete the sample construction based on the high-quality scheduling scheme.Then,ReliefF algorithm was adopted to filter redundant production features and obtain scheduling samples for training and prediction.Moreover,the probabilistic neural network combined with whale optimization algorithm was used as the decision-making model to realize the training and prediction process based on scheduling samples.Finally,the experimental results showed that the proposed method had good feature selection ability and high prediction accuracy.Compared with other real-time scheduling methods,it had better performance,and could effectively guide the manufacturing execution process according to the real-time state of shop floor.

Key words: hybrid flow shop, dynamic scheduling, production feature selection, probabilistic neural network, whale optimization algorithm

摘要： 在智能制造背景下,物联网等信息技术为制造系统积累了大量数据,同时人工智能等先进方法为车间数据分析和实时控制提供了有效手段。因此,针对不相关并行机混合流水车间调度问题,提出了一种基于生产数据的动态调度方法,以实现订单完工时间最小化。首先以高质量调度方案为基础,从中提取生产特征和调度规则完成样本构建。其次使用ReliefF算法过滤冗余生产特征,获得用于训练和预测的调度样本。然后采用融合鲸鱼优化算法的概率神经网络作为调度模型,实现基于调度样本的训练和预测过程。最后,实验结果表明,所提方法具有良好的特征选择能力和较高的预测精度,与其他实时调度方法相比具有更加优越的性能,可以有效地根据车间实时状态指导制造执行过程。

关键词: 混合流水车间, 动态调度, 生产特征选择, 概率神经网络, 鲸鱼优化算法