基于机理模型与深度学习的密炼工艺预测控制方法
Predictive Control Method of Mixing Process Based on Mechanism Model and Deep Learning
关瑞琦 1苏永清 1白雪杨 1崔哲昕1
作者信息
- 1. 同济大学 电子与信息工程学院,上海 201804
- 折叠
摘要
针对轮胎生产中密炼工艺的时变特性以及生产多要素对过程控制的要求,采用键合图和深度学习算法设计了一种模型预测控制方法,形成了对密炼过程生产多要素的有效控制.通过状态空间方程模型加反馈校正的方式,减小线性预测模型与非线性过程间误差;通过卷积神经网络(CNN)提取密炼工艺过程数据特征,通过长短期神经网络(LSTM)提取数据前后的时序关系;提出基于CNN-LSTM网络结构的系统参数辨识方法,实时更新状态空间预测模型参数,实现了预测模型与密炼过程的精准匹配.仿真结果验证了所提方法的有效性.
Abstract
Aiming at the time-varying characteristics of the mixing process in tire production and the requirement of multi-factor process control,a model predictive control method is proposed by using bond graph and deep learning algorithm,which formed an effective control of multi-factor production in the mixing process.The error between linear prediction model and nonlinear process is reduced through feedback correction.The feature extraction is carried out on the mixing process data sequence through CNN algorithm,and long and short term neural network(LSTM)is used to extract the timing characteristics of the data.The system parameter identification method based on CNN-LSTM network structure is proposed,and the parameters of the state-space prediction model are updated in real-time,which realizes the accurate matching between the prediction model and the mixing process.Simulation results verify the ef-fectiveness of the proposed method.
关键词
预测控制/密炼工艺/键合图/CNN-LSTMKey words
predictive control/mixing process/bond graph/CNN-LSTM引用本文复制引用
基金项目
国家重点研发计划项目(2022YFB3305300)
出版年
2024
NETL
NSTL
维普
万方数据