摘要
针对冷连轧机振动具有非线性、非平稳,以及与当前和历史状态息息相关的特点,提出了基于集合经验模态分解(EEMD)-长短时记忆循环神经网络(LSTM)的轧机振动预测模型。采用EEMD方法将轧机振动加速度分解为若干个频率单一、相对平稳的IMF模态分量和残差分量,有效地降低了振动加速度信号的复杂性;采用具有记忆单元的LSTM网络建立轧机振动预测模型,并通过引入历史振动信息显著提高了轧机振动的预测精度。仿真结果表明,EEMD-LSTM模型较LSTM模型的预测精度提高了11%,对轧机振动有很好的预测效果,并分析了各工艺参数与轧机振动之间的定量关系,为快速抑制轧机振动、优化轧制规程提供了参考。
For the characteristics of non-linear and non-stationary for vibration of tandem cold rolling mill, and it is closely related to the current and historical states, based on Ensemble Empirical Mode Decomposition(EEMD)-Long and Short Term Memory Recurrent Neural Network(LSTM), a rolling mill vibration prediction model was proposed. Then, the rolling mill vibration acceleration was decomposed into several IMF modal components and residual components with single frequency and relative stability by the EEMD method, and the complexity of vibration acceleration signal was reduced effectively. Furthermore, the prediction model of rolling mill vibration was established by using LSTM network with memory unit, and the prediction accuracy of rolling mill vibration was significantly improved by introducing historical vibration information. The simulation results show that the prediction accuracy of EEMD-LSTM model is 11% higher than that of LSTM model, and it has a good prediction effect on the rolling mill vibration. Meanwhile, the quantitative relationship between each process parameter and rolling mill vibration is analyzed, which provides a reference for quickly suppressing the rolling mill vibration and optimizing the rolling schedule.
作者
张瑞成
曹志新
Zhang Ruicheng;Cao Zhixin(School of Electrical Engineering,North China University of Science and Technology,Tangshan 063210,China)
出处
《锻压技术》
CAS
CSCD
北大核心
2022年第9期174-181,共8页
Forging & Stamping Technology
基金
河北省自然科学基金资助项目(F2018209201)。
关键词
冷连轧
轧机振动预测
EEMD分解
LSTM网络
振动加速度
tandem cold rolling
vibration prediction of rolling mill
EEMD decomposition
LSTM network
vibration acceleration
