基于数据驱动的断纸监测算法研究

Paper Break Monitoring Algorithm Based on Data-Driven Model

摘要: 为对造纸厂易断纸的突发情况进行有效预测，确保机器断纸前及时进行有效预警，减少断纸次数，节约成本，本文将对五种基于数据驱动的断纸监测方法进行研究。为监测生产过程中的断纸情况，模型采用相关系数法除去冗余变量和无关变量，并通过点二列相关法筛选出核心特征变量。将数据分别输入到随机森林、LSTM、SVM、Transform和LSTM-RF五种算法中进行对比实验。本文利用国内某造纸厂造纸系统的实时生产数据对模型进行验证，结果表明，选择SVM和LSTM-RF模型其他算法具有更高稳定性，且监测造纸过程中断纸情况的准确度较其他几种算法更高，起到更好的预测作用。

Abstract: To effectively predict the sudden situation of paper breakage in paper mills, ensure timely and effective early warning before machine paper breakage, reduce the frequency of paper breakage and save costs, this paper will study five data-driven paper breakage monitoring methods. To monitor the paper breakage situation during the production process, the model adopts the correlation coefficient method to eliminate redundant and irrelevant variables, and screens out the core characteristic variables through the point-two-column correlation method. The data were respectively input into five algorithms, namely Random Forest, LSTM, SVM, Transform and LSTM-RF, for comparative experiments. In this paper, the real-time production data of the papermaking system of a domestic paper mill was used to verify the model. The results show that the selection of SVM and other algorithms of the LSTM-RF model has higher stability, and the accuracy of monitoring the situation of interrupted paper in the papermaking process is higher than that of several other algorithms, playing a better predictive role.