基于随机卷积核神经网络数据增强的软测量  

Soft Sensor Based on Random Convolutional Kernel Neural Network Data Enhancement

作　　者：钱慧 刘瑞兰 QIAN Hui;LIU Ruilan(College of Automation&College of Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学自动化学院、人工智能学院,江苏南京210023

出　　处：《软件导刊》2024年第6期53-58,共6页Software Guide

摘　　要：精对苯二甲酸(PTA)生产过程中PX氧化反应的副产品4-CBA难以在线测量,只能通过离线分析获得少量样本。针对该问题,提出一种基于随机卷积核神经网络数据增强的动态软测量模型RCKN-XGBoost。该模型首先采用随机卷积核神经网络(RCKN)进行数据增强,扩充样本数量并改善其多样性;然后将原始样本与扩充后的样本线性组合,构成新的数据集;最后采用XGBoost对网络进行训练与预测。在某化工厂PX氧化过程4-CBA含量数据集上对RCKN-XGBoost模型与XGBoost、CNN、CNN-XGBoost、Laplace-XGBoost模型进行比较,发现RCKN-XGBoost模型的MRE指标分别降低了1.07%、0.92%、0.80%和0.65%,RMSE分别降低了27.9、18.62、12.58和8.05,证明了该模型的有效性。The by-product 4-CBA of PX oxidation reaction in the production process of purified terephthalic acid(PTA)is difficult to mea-sure online,and only a small amount of samples can be obtained through offline analysis.A dynamic soft sensing model RCKN-XGBoost based on random convolutional kernel neural network data augmentation is proposed to address this issue.The model first uses random convolu-tional kernel neural network(RCKN)for data augmentation,expanding the sample size and improving its diversity;Then,the original sample is linearly combined with the expanded sample to form a new dataset;Finally,XGBoost was used to train and predict the network.On the 4-CBA content dataset of PX oxidation process in a certain chemical plant,the RCKN-XGBoost model was compared with XGBoost,CNN,CNN-XGBoost,and Laplace XGBoost models.It was found that the MRE index of the RCKN-XGBoost model decreased by 1.07%,0.92%,0.80%,and 0.65%,respectively,and the RMSE decreased by 27.9%,18.62%,12.58%,and 8.05%,proving the effectiveness of the model.

关 键 词：软测量 4-CBA 随机卷积核神经网络 数据增强 XGBoost 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]