安全、健康和环境2025，Vol.25Issue(6)：40-48,9.DOI:10.3969/j.issn.1672-7932.2025.06.006
基于改进的LSTM网络的石油化工传感器故障预测研究
Research on Fault Prediction of Petrochemical Sensors Based on the Improved LSTM Network
摘要
Abstract
In petrochemical production facilities,temperature and pressure sensors are in high-temperature and high-pressure working conditions for a long time.The early characteristics of faults are highly concealed,and the existing methods are difficult to achieve high-precision advanced early warning.Aiming at the problem of difficult early warning of faults in petrochemical sensors,an improved LSTM network architecture was pro-posed.By optimizing the LSTM network structure,the cell state memory function is integratd the forgetting gate mechanism,enhancing the temporal correlation and synergistic characteristics of sensor data.The sensor data of the oil refining unit were collected at the industrial site to construct the training set and conduct predictive per-formance tests under different working conditions.The results showed that the advanced 35-min prediction accu-racy of the improved LSTM model for temperature and pressure sensors reached 99.75％and 99.68％,respec-tively.In the case analysis of the acid gas absorption device,the false alarm rate was lower than 0.12％,and the missed alarm rate remained stable within 10.20％.The research found that when the sensor presented pa-rameter coupling characteristics and had a high consistency in the time series,the model had a more acute abili-ty to capture fault characteristics and can predict the early fault characteristics of the sensor in a high-tempera-ture and high-pressure environment.This research provided an effective technical means for the fault prediction of sensors in petrochemical processes.关键词
深度强化学习/传感器/石油化工/故障预测/LSTM网络Key words
deep reinforcement learning/sensor/petrochemical industry/fault prediction/LSTM network分类
信息技术与安全科学引用本文复制引用
李娜..基于改进的LSTM网络的石油化工传感器故障预测研究[J].安全、健康和环境,2025,25(6):40-48,9.基金项目
国家重点研发计划项目(2024YFC3013603),全流程控制系统性能退化智能评估和安全性能自主提升技术及设备. （2024YFC3013603）
