【作者】 罗顺桦； 王振雷； 王昕；

【Author】 LUO Shun-hua;WANG Zhen-lei;WANG Xin;Key Laboratory of Advanced Control and Optimization for Chemical Processes, East China University of Science and Technology;Center of Electrical & Electronic Technology,Shanghai Jiao Tong University;

【机构】 华东理工大学化工过程先进控制和优化技术教育部重点实验室； 上海交通大学电工与电子技术中心；

【摘要】 软测量建模通过选取辅助变量,建立辅助变量与关键质量变量关系,能够高效地实现对关键质量变量的预测。然而当辅助变量维数较高,且对关键质量变量的影响程度不一时,将导致网络预测误差较大。针对这一问题,本文提出一种基于注意力机制的multi-headCNN-LSTM模型,首先根据辅助变量自身属性和特点将其切分成多组子变量后,使用多组独立并行工作的CNN-LSTM群对其子变量进行单独处理,再提取各个子变量上的特征向量,融合注意力机制,实现子变量特征向量的权重分配。本文所提算法不需提前根据工艺知识选择辅助变量,而是通过深度学习机制自动选择特征。最后,在乙烯精馏塔塔顶乙烷浓度软测量建模中进行应用,本文模型的预测精度优于LSTM以及CNN-LSTM软测量模型。更多还原

【Abstract】 By selecting auxiliary variables and establishing the relationship between the auxiliary variables and key quality variables, soft sensing modeling can effectively predict the key quality variables. However, because of the high dimension of the auxiliary variables and its different infulence on the key quality variables, it will result in the large network prediction errors. Aiming at this problem, we propose a multi-head CNN-LSTM model based on attention mechanism. Firstly, according to auxiliary variables’ attributes and characteristics, we divide them into multiple groups of sub-variables and use multiple groups of CNN-LSTM working independently and in parallel to process their sub-variables. Then by extracting feature vectors on each sub-variable, we combine attention mechanism to achieve weight distribution of feature variables of subvariables. The algorithm proposed in this paper does not need to select auxiliary variables according to process knowledge in advance, but automatically selects features through a deep learning mechanism. Finally, it is applied in the soft sensing modeling of ethane concentration on the top of ethylene distillation tower. The result shows that the model we propose performs better than LSTM and CNN-LSTM soft sensing models更多还原