基于动态残差补偿的工业时序数据故障检测模型

doi:10.11949/0438-1157.20251250

摘要/Abstract

摘要：

为应对化工过程因动态特性、强非线性与复杂时空耦合导致传统故障检测模型精度低且动态适应性差的难题，提出一种基于动态残差补偿的工业时序数据故障检测模型（Dynamic Residual Compensation-based Detection Model， DRCDM）。DRCDM利用卷积神经网络-双向长短期记忆网络（CNN-BiLSTM）挖掘时序数据的时空依赖关系并进行初步预测，随后构建包含原始数据统计量、基础模型预测值和历史残差的动态统计量的增强元特征集，再驱动XGBoost对预测偏差实时建模与修正。在田纳西-伊斯曼（TE）过程及某280万吨/年催化裂化（FCC）装置运行数据的实验表明，动态残差补偿能够有效实现偏差的实时修正，显著提升DRCDM故障检测的准确性和可靠性。

关键词: 故障检测, 动态残差补偿, 时序数据, 元特征集, TE过程, FCC

Abstract:

This paper addresses the challenge that traditional fault detection models fail to simultaneously achieve high precision and dynamic adaptability due to the inherent dynamic characteristics, strong nonlinearity, and complex spatiotemporal coupling of chemical processes. Therefore, we propose a dynamic residual compensation-based detection model for fault detection in industrial time-series data. The proposed DRCDM operates by first capturing the spatio-temporal dependencies within time series data to generate preliminary predictions. Subsequently, it constructs an enhanced meta-feature set that incorporates raw data statistics, predictions from the base model, and dynamic historical residual statistics to drive XGBoost for real-time modeling and correction of prediction deviations. The effectiveness of DRCDM is validated using the Tennessee-Eastman (TE) process dataset, along with industrial operational data from a 2.8×10⁶ tons/year fluid catalytic cracking (FCC) unit, with performance comparisons made against several established methods. The results demonstrate that dynamic residual compensation achieves real-time deviation correction, thereby significantly improving the accuracy and reliability of fault detection in DRCDM.

Key words: fault detection, dynamic residual compensation, time series data, meta-feature set, TE process, FCC

中图分类号:

TP183

吉鲁东, 朱春梦, 柳楠, 郭建豪, 赵云鹏, 石孝刚, 蓝兴英. 基于动态残差补偿的工业时序数据故障检测模型[J]. 化工学报, DOI: 10.11949/0438-1157.20251250.

Ludong JI, Chunmeng ZHU, Nan LIU, Jianhao GUO, Yunpeng ZHAO, Xiaogang SHI, Xingying LAN. Industrial time series data fault detection model based on dynamic residual correction[J]. CIESC Journal, DOI: 10.11949/0438-1157.20251250.

图/表 23

参考文献 32

[1]	Bai Y M, Xiang S Y, Cheng F F, et al. A dynamic-inner LSTM prediction method for key alarm variables forecasting in chemical process[J]. Chinese Journal of Chemical Engineering, 2023, 55: 266-276.
[2]	Xiang S Y, Bai Y M, Zhao J S. Medium-term prediction of key chemical process parameter trend with small data[J]. Chemical Engineering Science, 2022, 249: 117361.
[3]	Jia F, Lei Y G, Guo L, et al. A neural network constructed by deep learning technique and its application to intelligent fault diagnosis of machines[J]. Neurocomputing, 2018, 272: 619-628.
[4]	冯子芸, 王治红, 戴一阳. 基于随机森林的不可靠数据化工过程故障诊断方法[J]. 石油学报(石油加工), 2021, 37(1): 121-129.
	Feng Z Y, Wang Z H, Dai Y Y. Random forest based fault diagnosis method for chemical process with unreliable data[J]. Acta Petrolei Sinica (Petroleum Processing Section), 2021, 37(1): 121-129.
[5]	Wang Q L, Han J Q, Chen F, et al. Real-time risk prediction of chemical processes based on attention-based Bi-LSTM[J]. Chinese Journal of Chemical Engineering, 2024, 75: 131-141.
[6]	杨余, 杨鑫, 王英, 等. 基于mini-1D-CNN模型的TE过程故障诊断[J]. 中国安全科学学报, 2023, 33(2): 173-178.
	Yang Y, Yang X, Wang Y, et al. Research on TE process fault diagnosis based on mini-1D-CNN model[J]. China Safety Science Journal, 2023, 33(2): 173-178.
[7]	张坤, 肖慧, 徐哈宁, 等. 基于主成分分析和长短期记忆网络的滑坡地表位移监测数据缺失插补算法[J]. 科学技术与工程, 2023, 23(26): 11129-11135.
	Zhang K, Xiao H, Xu H N, et al. Missing interpolation algorithm of landslide surface displacement monitoring data based on principal component analysis and long short term memory network[J]. Science Technology and Engineering, 2023, 23(26): 11129-11135.
[8]	孔祥玉, 陈雅琳, 罗家宇, 等. 基于偏最小二乘的多特性复杂过程监测方法[J]. 华南理工大学学报(自然科学版), 2022, 50(6): 100-110.
	Kong X Y, Chen Y L, Luo J Y, et al. Multi-feature complex process monitoring method based on partial least squares[J]. Journal of South China University of Technology (Natural Science Edition), 2022, 50(6): 100-110.
[9]	曹岚, 周成功, 袁斌霞, 等. 基于梯度提升回归树-自训练贝叶斯优化支持向量机的光伏阵列故障诊断方法[J]. 太阳能学报, 2025, 46(6): 289-297.
	Cao L, Zhou C G, Yuan B X, et al. Fault diagnosis methods of photovoltaic arrays based on gradient boosting regression tree-self-training Bayesian optimized support vector machine[J]. Acta Energiae Solaris Sinica, 2025, 46(6): 289-297.
[10]	严爱军, 王英杰, 王殿辉. Tennessee-Eastman过程的学习型案例推理故障诊断方法[J]. 控制理论与应用, 2017, 34(9): 1179-1184.
	Yan A J, Wang Y J, Wang D H. Fault diagnosis method using learning case-based reasoning for Tennessee-Eastman process[J]. Control Theory & Applications, 2017, 34(9): 1179-1184.
[11]	Perera Y S, Ratnaweera D A A C, Dasanayaka C H, et al. The role of artificial intelligence-driven soft sensors in advanced sustainable process industries: a critical review[J]. Engineering Applications of Artificial Intelligence, 2023, 121: 105988.
[12]	Liu N, Zhu C M, Li Y H, et al. Hybrid deep learning framework with spatiotemporal pattern extraction for decant oil solid content soft sensor development in fluid catalytic cracking units[J]. Petroleum Science, 2025, 22(7): 3042-3055.
[13]	Zhang Z P, Zhao J S. A deep belief network based fault diagnosis model for complex chemical processes[J]. Computers & Chemical Engineering, 2017, 107: 395-407.
[14]	Xu Y Y, Li Y, Wang Y J, et al. Improved few-shot learning method for transformer fault diagnosis based on approximation space and belief functions[J]. Expert Systems with Applications, 2021, 167: 114105.
[15]	Guidotti R, Monreale A, Ruggieri S, et al. A survey of methods for explaining black box models[J]. ACM Computing Surveys, 2019, 51(5): 1-42.
[16]	Zhou Z, Yu H W, Yang L, et al. Fault diagnosis method for chemical processes based on variable correlation-guided convolutional neural networks[J]. The Canadian Journal of Chemical Engineering, 2025, 103(10): 4874-4889.
[17]	许洪光, 李凤英, 郭茜. 机器学习算法改进及在化工故障诊断中的应用[J]. 粘接, 2022, 49(5): 85-89, 94.
	Xu H G, Li F Y, Guo Q /X). Improvement of machine learning algorithm and its application in chemical fault diagnosis[J]. Adhesion, 2022, 49(5): 85-89, 94.
[18]	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.
	Li N. Research on fault prediction of petrochemical sensors based on the improved LSTM network[J]. Safety Health & Environment, 2025, 25(6): 40-48.
[19]	Sun L, You F Q. Machine learning and data-driven techniques for the control of smart power generation systems: an uncertainty handling perspective[J]. Engineering, 2021, 7(9): 1239-1247.
[20]	Bayram F, Ahmed B S, Kassler A. From concept drift to model degradation: an overview on performance-aware drift detectors[J]. Knowledge-Based Systems, 2022, 245: 108632.
[21]	Gorinevsky D. Fault isolation in data-driven multivariate process monitoring[J]. IEEE Transactions on Control Systems Technology, 2015, 23(5): 1840-1852.
[22]	Lu C, Fei J Y, Meng X Z, et al. Thermal error prediction and compensation of digital twin laser cutting based on T-XGBoost[J]. Sensors, 2022, 22(18): 7022.
[23]	魏国家, 赵冠, 赵巍, 等. 数字孪生框架下五轴铣削轮廓误差预测与补偿研究[J]. 机械设计与制造, 2025, 10(10): 1-9.
	Wei G J, Zhao G, Zhao W, et al. Experimental Study on Contour Error Prediction and Dynamic Compensation Method for Five-Axis Milling Based on Digital Twin[J]. Machinery Design & Manufacture, 2025, 10(10): 1-9. [24] Thoma S. Model based compensation of dynamic path deviations in 5-axis machine tools[J]. 2015.
[25]	Li T Y, Hua M, Wu X. A hybrid CNN-LSTM model for forecasting particulate matter (PM_2.5)[J]. IEEE Access, 2020, 8: 26933-26940.
[26]	谢世元, 郭志远, 强惠敏, 等. 基于XGBoost算法的船舶目的地预测模型[J]. 计算机工程与设计, 2025, 46(6): 1766-1771.
	Xie S Y, Guo Z Y, Qiang H M, et al. Vessel destination prediction model based on XGBoost algorithm[J]. Computer Engineering and Design, 2025, 46(6): 1766-1771.
[27]	宋贽, 王晗, 张久军. 网络投诉反馈视角下汽车质量缺陷的动态监测与预警研究[J]. 运筹与管理, 2025: 1-8.
	Song Z, Wang H, Zhang J J. Dynamic Monitoring and Early Warning Research on Automobile Quality Defects from the Perspective of Online Complaint Feedback[J]. Operations Research and Management Science, 2025: 1-8.
[28]	姜旭峰, 陈勇旭, 宋宝宝, 等. 基于油液在线监测数据的海警船状态异常报警策略研究[J]. 润滑与密封, 2025: 1-11.
	Jiang X F, Chen Y X, Song B B, et al. Research on abnormal state alarm strategy of coast guard ship based on oil online monitoring data [J]. Lubrication Engineering, 2025: 1-11.
[29]	Downs J J, Vogel E F. A plant-wide industrial process control problem[J]. Computers & Chemical Engineering, 1993, 17(3): 245-255.
[30]	Bai Y M, Zhao J S. A novel transformer-based multi-variable multi-step prediction method for chemical process fault prognosis[J]. Process Safety and Environmental Protection, 2023, 169: 937-947.
[31]	Bi X T, Zhao J S. A novel orthogonal self-attentive variational autoencoder method for interpretable chemical process fault detection and identification[J]. Process Safety and Environmental Protection, 2021, 156: 581-597.
[32]	Bathelt A, Ricker N L, Jelali M. Revision of the Tennessee Eastman process model[J]. IFAC-PapersOnLine, 2015, 48(8): 309-314.

Model	Parameter
CNN	(Conv1D) Filter=128, Kernel=3, Activation=Relu
	Dropout=0.2
	(Conv1D) Filter=64, Kernel=3, Activation=Relu
BiLSTM	BiLSTM=64
	Dropout=0.2
XGBoost	N_estimators=[100, 200, 300]
	Max_depth=[3, 5, 7]
	Learning_rate=[0.01, 0.05, 0.1]
	Subsampl=[0.8, 1.0]
	Colsample_bytree=[0.8, 1.0]

Model	Parameter
CNN	(Conv1D) Filter=128, Kernel=3, Activation=Relu
	Dropout=0.2
	(Conv1D) Filter=64, Kernel=3, Activation=Relu
BiLSTM	BiLSTM=64
	Dropout=0.2
XGBoost	N_estimators=[100, 200, 300]
	Max_depth=[3, 5, 7]
	Learning_rate=[0.01, 0.05, 0.1]
	Subsampl=[0.8, 1.0]
	Colsample_bytree=[0.8, 1.0]

故障编号	故障描述	故障类型
IDV1	A/C进料的比率，B成分不变（流4）	阶跃
IDV2	B成分变化，A/C进料比率不变（流4）	阶跃
IDV3	D的进料温度变化（流2）	阶跃
IDV4	反应器冷却水入口温度变化	阶跃
IDV5	冷凝器冷却水入口温度变化	阶跃
IDV6	A进料损失（流1）	阶跃
IDV7	C存在压力损失-可用性降低（流4）	阶跃
IDV8	A、B、C进料成分变化（流4）	随机变量
IDV9	D的进料温度变化（流2）	随机变量
IDV10	C的进料温度变化（流4）	随机变量
IDV11	反应器冷却水入口温度变化	随机变量
IDV12	冷凝器冷却水入口温度变化	随机变量
IDV13	反应动力学变化	缓慢偏移
IDV14	反应器冷却水阀门	阀粘滞
IDV15	冷凝器冷却水阀门	阀粘滞

故障编号	故障描述	故障类型
IDV1	A/C进料的比率，B成分不变（流4）	阶跃
IDV2	B成分变化，A/C进料比率不变（流4）	阶跃
IDV3	D的进料温度变化（流2）	阶跃
IDV4	反应器冷却水入口温度变化	阶跃
IDV5	冷凝器冷却水入口温度变化	阶跃
IDV6	A进料损失（流1）	阶跃
IDV7	C存在压力损失-可用性降低（流4）	阶跃
IDV8	A、B、C进料成分变化（流4）	随机变量
IDV9	D的进料温度变化（流2）	随机变量
IDV10	C的进料温度变化（流4）	随机变量
IDV11	反应器冷却水入口温度变化	随机变量
IDV12	冷凝器冷却水入口温度变化	随机变量
IDV13	反应动力学变化	缓慢偏移
IDV14	反应器冷却水阀门	阀粘滞
IDV15	冷凝器冷却水阀门	阀粘滞

编号	运行时长/h	样本数量
IDV0（正常）	24.0	480
IDV6	14.1	282
IDV1-5,7-15	48.0	960