Abstract: To predict the optimal scheduling mode under different failure scenarios, and achieve intelligent and effective scheduling decisions, an rescheduling mode prediction method was proposed integrating scheduling simulation and improved PNN. Firstly, in face of difficulty in sample collection and incompleteness in failure samples, the large-scale samples with randomness were generated on the ground of simulation methods. Moreover, after collecting the data about the cumulative changes of processing time, the numbers of changed operations and the changes of makespan, the label of each sample were generated. Particularly, all the tagged data were put into the PNN model, and the optimal rescheduling mode was predicted. The experimental results demonstrate that the accuracy rate of the proposed method reaches over 99.54%. On the premise of the specified processing workshop and production tasks, the failure machine and repair duration play a decisive role in the optimal rescheduling mode.

Key words: machine failure, job shop, sample simulation, rescheduling mode prediction, probabilistic neural network(PNN)

摘要： 为预知不同故障情形下的优化重调度方式，实现快速、有效的重调度决策，提出融合调度仿真与改进概率神经网络的重调度方式预测方法。考虑到现场故障样本难获得且无法涵盖全部故障情形，利用仿真实现随机故障下优化重调度样本的生成；以工序加工时间的累计变动、变动任务数、makespan改变量为决策依据，生成各样本的标签；将带标签数据样本输入到概率神经网络模型，实现优化重调度方式预测。实验结果表明：所提出的方法准确率达99.54%；在指定加工车间和生产任务的前提下，故障机序号和故障修复时间对优化重调度方式起决定性作用。

关键词: 机器故障, 加工车间, 样本仿真, 重调度方式预测, 概率神经网络