可重入制造系统的产出短期预测方法研究

The short-term prediction method of throughput for En-entrant manufacturing system

针对可重入制造系统处于高度不确定性环境中,其产出时间序列的非线性特征,提出了基于遗传小波神经网络的产出短期预测方法,使预测模型具有小波的优良逼近性质和神经网络的自学习自适应性质,同时,采用遗传算法训练神经网络参数,利用遗传算法隐含并行性和全局搜索性的特征,丰富优化过程中的搜索行为,增强全局和局部意义下的搜索能力和效率。通过半导体生产线实例,进行了验证,结果表明所提出的基于遗传小波神经网络的产出短期预测方法的预测性能要优于传统BP神经网络算法。

The re-entrant manufacturing systems （RMS） under the high uncertain environment have the non-linear characteristic of time series in the throughput. GA ＆ wavelet neural network based short-term prediction method of throughput is proposed. It makes the prediction model have the excellent toward feature of wavelet analysis and the self-study ＆ self-adaptive characteristic of neural network. At the same time, the parameters of neural network are trained by the genetic algorithm, and the search behaviors of optimal process are enriched by using the characteristic of recessive concurrent and whole search of genetic algorithm so as to enhance the overall and local search capacity and efficiency. The Case study of semiconductor production line is as experiment. The experimental results demonstrate the effectiveness of the above proposed method and it has precedence over the back propagation （BP） of neural network in the prediction performance.