摘要
为了提升页岩炼油厂关键设备的故障预警能力,以某典型项目中的尾气压缩机、空压机和燃烧风机为研究对象,提出了一种改进型长短时记忆(LSTM)神经网络模型。通过引入注意力机制优化网络结构,结合自适应学习率调整策略和数据增强技术,提升了模型的预测精度与训练效率。实验结果表明,系统在3类设备上的平均准确率达到94.3%,平均预警提前时间为48.2 min,误报率为6.4%。该系统有效解决了传统LSTM训练慢、易陷入局部最优的问题,具备良好的工程应用前景。
To improve the fault early warning capability of key equipment in shale oil refineries,this paper takes the tail gas compressor,air compressor and combustion fan in a typical project as the research objects and proposes an improved Long Short-Term Memory(LSTM)neural network model.By introducing the attention mechanism to optimize the network structure,combined with the adaptive learning rate adjustment strategy and data augmentation technology,the prediction accuracy and training efficiency of the model are improved.Experimental results show that the average accuracy of the system on the three types of equipment reaches 94.3%,the average early warning lead time is 48.2 minutes,and the false alarm rate is 6.4%.The system effectively solves the problems of slow training and easy trapping in local optimum of the traditional LSTM,and has good prospects for engineering application.
作者
王笃鹏
闫玉麟
杨勇
WANG Dupeng;YAN Yulin;YANG Yong(Shengli Experimental Factory of Fushun Mining Group Co.,Ltd.,Fushun,Liaoning 113000,China)
出处
《自动化应用》
2026年第2期42-44,47,共4页
Automation Application
