基于集成学习传递熵的化工过程微小故障检测方法

doi:10.11949/0438-1157.20230287

摘要/Abstract

摘要：

传统的多元统计分析方法直接对过程数据的均值和方差进行故障分析，但只要监测统计量保持在控制极限所包围的正态区域内，就不能检测出数据分布的变化。针对这一问题，提出一种基于集成学习传递熵的微小故障检测方法。该方法首先利用传递熵提取变量间的信息传递，通过滑动窗口获取传递熵数据集，并根据此数据集求取监测统计量和控制限。然后，利用秩和比法对构建的评价指标进行排序分档，筛选出分档结果良好的传递熵数据集对应的监测统计量。最后，利用集成学习与贝叶斯推理策略相结合的算法对监测结果进行融合，实时检测过程故障。将该方法用于数值例子和连续搅拌反应釜过程的故障检测，并与核主元分析、核独立元分析、加权统计局部核主元分析和加权统计特征核独立元分析进行对比，验证了所提方法具有良好的微小故障检测性能。

关键词: 微小故障, 主元分析, 传递熵, 秩和比法, 集成学习

Abstract:

Traditional multivariate statistical analysis methods directly perform fault analysis on the mean and variance of process data. But as long as the monitoring statistics remain within the normal region surrounded by control limits, changes in the data distribution cannot be detected. To solve this problem, a method based on ensemble learning transfer entropy is proposed to obtain monitoring statistics and control limits based on this data set. Firstly, the transfer entropy is used to measure the information transfer between variables, and then the transfer entropy dataset is obtained through a sliding window, and the monitoring statistics and control limits are derived from this dataset. Then, the constructed indicators are ranked and binned using the rank sum ratio method, and the probability density parameters of the transfer entropy and the corresponding monitoring statistics of the part with good binning results are screened. Finally, the monitoring results are fused using an algorithm combining ensemble learning and Bayesian inference strategies to detect process faults in real-time. The method is used for fault detection of numerical examples and continuously stirred tank reactor processes and is compared with kernel principal component analysis, kernel independent component analysis, weighted statistical local kernel principal component analysis and weighted statistical feature kernel independent component analysis to verify that the proposed the method has good performance in detecting minor faults.

Key words: incipient fault, principal component analysis, transfer entropy, rank sum ratio method, ensemble learning

中图分类号:

TP 277

王光, 单发顺, 钱禹丞, 焦建芳. 基于集成学习传递熵的化工过程微小故障检测方法[J]. 化工学报, 2023, 74(7): 2967-2978.

Guang WANG, Fashun SHAN, Yucheng QIAN, Jianfang JIAO. Incipient fault detection method for chemical process based on ensemble learning transfer entropy[J]. CIESC Journal, 2023, 74(7): 2967-2978.

图/表 13

图1 基于ETEn的故障监测流程图

Fig.1 Flow chart of ETEn-based fault monitoring

表1 评价指标成对比较矩阵及权重

Table 1 Pairwise comparison matrix and weights of evaluation index

指标	ACC	TPR	FPR	TNR	FNR	Precision	F	权值
ACC	1	5	5	5	5	5	3	0.412
TPR	1/5	1	1	1	1	1	1/3	0.076
FPR	1/5	1	1	1	1	1	1/3	0.076
TNR	1/5	1	1	1	1	1	1/3	0.076
FNR	1/5	1	1	1	1	1	1/3	0.076
Precision	1/5	1	1	1	1	1	1/3	0.076
F	1/3	3	3	3	3	3	1	0.211

表2 数值例子参数B的RSR分档结果

Table 2 RSR binning results for parameter B in the numerical example

分档	概率单位Z	WRSR	分档结果^②
$R S R u p$ ^①	>6	>0.721	39、34、29、35
$R S R u p$ ^①	>6	>0.724	35、40、31、33
$R S R m i d$	4~6	0.273~0.721	28、21、25、27、30、26、32、38、40、36、31、33、20、37
$R S R m i d$	4~6	0.274~0.724	39、34、24、32、30、37、29、20、36、38、27、23、26、22
$R S R d o w n$	<4	<0.273	22、24、23
$R S R d o w n$	<4	<0.274	28、25、21

表2 数值例子参数B的RSR分档结果

Table 2 RSR binning results for parameter B in the numerical example

分档	概率单位Z	WRSR	分档结果^②
$R S R u p$ ^①	>6	>0.721	39、34、29、35
$R S R u p$ ^①	>6	>0.724	35、40、31、33
$R S R m i d$	4~6	0.273~0.721	28、21、25、27、30、26、32、38、40、36、31、33、20、37
$R S R m i d$	4~6	0.274~0.724	39、34、24、32、30、37、29、20、36、38、27、23、26、22
$R S R d o w n$	<4	<0.273	22、24、23
$R S R d o w n$	<4	<0.274	28、25、21

图2 不同传递熵窗口参数对应的故障误报率

Fig.2 The fault alarm rate corresponding to different transfer entropy window parameters

图3 数值例子故障1的监控图

Fig.3 Fault 1 monitoring diagrams of the numerical examples

表3 五种方法对于数值例子的检测性能

Table 3 The detection performance of the five methods for numerical examples

故障		KPCA		KICA		WSLKPCA		WSFKICA		ETEn
故障		$T 2$	$S P E$	$I 2$	$S P E$	$T 2$	$S P E$	$T 2$	$S P E$	$E T 2$	$E S P E$
1	故障检测率	54.0%	0	5.4%	81.8%	30.9%	97.0%	68.9%	82.9%	93.0%	99.0%
	误报率	0.6%	0	0.7%	0.9%	9.1%	4.2%	3.8%	2.4%	0	0
	DD/min	57	ND	ND	6	150	10	38	54	9	9
2	故障检测率	60.8%	0	13.8%	66.4%	84.4%	68.3%	69.9%	76.0%	76.7%	80.6%
	误报率	2.0%	0	0.6%	0.7%	0.8%	5.1%	4.2%	1.6%	0	0
	DD/min	233	ND	ND	219	9	93	37	28	63	45
平均故障检测率		57.4%	0	9.6%	74.1%	57.7%	82.7%	69.4%	79.5%	84.9%	89.8%

表3 五种方法对于数值例子的检测性能

Table 3 The detection performance of the five methods for numerical examples

故障		KPCA		KICA		WSLKPCA		WSFKICA		ETEn
故障		$T 2$	$S P E$	$I 2$	$S P E$	$T 2$	$S P E$	$T 2$	$S P E$	$E T 2$	$E S P E$
1	故障检测率	54.0%	0	5.4%	81.8%	30.9%	97.0%	68.9%	82.9%	93.0%	99.0%
	误报率	0.6%	0	0.7%	0.9%	9.1%	4.2%	3.8%	2.4%	0	0
	DD/min	57	ND	ND	6	150	10	38	54	9	9
2	故障检测率	60.8%	0	13.8%	66.4%	84.4%	68.3%	69.9%	76.0%	76.7%	80.6%
	误报率	2.0%	0	0.6%	0.7%	0.8%	5.1%	4.2%	1.6%	0	0
	DD/min	233	ND	ND	219	9	93	37	28	63	45
平均故障检测率		57.4%	0	9.6%	74.1%	57.7%	82.7%	69.4%	79.5%	84.9%	89.8%

图4 闭环CSTR模型示意图

Fig.4 Schematic diagram of the closed-loop CSTR model

表4 闭环CSTR模型变量说明

Table 4 Description of the variables in the closed-loop CSTR model

索引	变量	说明
1	$C i$	进料浓度
2	$T i$	进料温度
3	$C$	反应物浓度
4	$T$	反应物温度
5	$Q c$	冷却剂流速
6	$T c i$	入口冷却剂温度
7	$T c$	冷却剂温度

表4 闭环CSTR模型变量说明

Table 4 Description of the variables in the closed-loop CSTR model

索引	变量	说明
1	$C i$	进料浓度
2	$T i$	进料温度
3	$C$	反应物浓度
4	$T$	反应物温度
5	$Q c$	冷却剂流速
6	$T c i$	入口冷却剂温度
7	$T c$	冷却剂温度

表5 CSTR系统故障

Table 5 CSTR system fault

故障	说明	$σ$
1	$C i = C i, 0 + σ$	0.06
2	$T i = T i, 0 + (k - 1000) σ$ ^①	1/3000
3	$T i = T i, 0 + σ ξ$ ^②	0.1

表5 CSTR系统故障

Table 5 CSTR system fault

故障	说明	$σ$
1	$C i = C i, 0 + σ$	0.06
2	$T i = T i, 0 + (k - 1000) σ$ ^①	1/3000
3	$T i = T i, 0 + σ ξ$ ^②	0.1

表6 CSTR系统参数B的RSR分档结果

Table 6 RSR binning results for parameter B in the CSTR system

分档	概率单位 $Z$	WRSR	分档结果
$R S R u p$	>6	>0.670	22、26、25、24
		>0.681	29、38、35、34
		>0.685	40、22、39、31
$R S R m i d$	4~6	0.319~0.670	32、31、27、37、34、36、33、28、23、29、35、39、20、38
		0.322~0.681	33、25、27、32、28、23、22、21、39、40、30、36、24、37
		0.323~0.685	25、29、20、38、30、33、27、34、37、36、26、24、28、21
$R S R d o w n$	<4	<0.319	21、40、30
		<0.322	31、20、26
		<0.323	32、35、23

表6 CSTR系统参数B的RSR分档结果

Table 6 RSR binning results for parameter B in the CSTR system

分档	概率单位 $Z$	WRSR	分档结果
$R S R u p$	>6	>0.670	22、26、25、24
		>0.681	29、38、35、34
		>0.685	40、22、39、31
$R S R m i d$	4~6	0.319~0.670	32、31、27、37、34、36、33、28、23、29、35、39、20、38
		0.322~0.681	33、25、27、32、28、23、22、21、39、40、30、36、24、37
		0.323~0.685	25、29、20、38、30、33、27、34、37、36、26、24、28、21
$R S R d o w n$	<4	<0.319	21、40、30
		<0.322	31、20、26
		<0.323	32、35、23

图5 CSTR系统故障1的监控图

Fig.5 Fault 1 monitoring diagram of the CSTR system

图6 CSTR系统故障3的监控图

Fig.6 Fault 3 monitoring diagram of the CSTR system

表7 五种方法对于CSTR系统的检测性能

Table 7 The detection performance of the five methods for the CSTR system

故障		KPCA		KICA		WSLKPCA		WSFKICA		ETEn
故障		$T 2$	SPE	$I 2$	SPE	$T 2$	SPE	$T 2$	SPE	$E T 2$	ESPE
1	故障检测率/%	57.6	0	45.8	39.0	93.4	76.1	84.2	93.6	99.4	99.0
	误报率/%	4.6	0	1.2	0.7	7.9	4.4	4.7	4.4	0	0
	DD/min	65	ND	42	46	38	113	0	0	8	10
2	故障检测率/%	48.4	0	10.6	36.4	68.9	72.7	28.9	76.1	99.4	86.6
	误报率/%	0.6	0	0.6	0.8	2.4	0.8	0	3.4	0.4	0
	DD/min	338	ND	447	294	161	142	80	8	10	74
3	故障检测率/%	38.2	0	18.6	20.6	83.4	76.1	69.3	63.5	99.6	99.2
	误报率/%	1	0	3.3	1.9	0	3.6	4.7	4.4	0.2	0
	DD/min	101	ND	93	ND	61	36	0	0	7	9
平均故障检测率/%		48.1	0	25.0	32.0	81.9	75.0	60.8	77.7	99.5	94.9

表7 五种方法对于CSTR系统的检测性能

Table 7 The detection performance of the five methods for the CSTR system

故障		KPCA		KICA		WSLKPCA		WSFKICA		ETEn
故障		$T 2$	SPE	$I 2$	SPE	$T 2$	SPE	$T 2$	SPE	$E T 2$	ESPE
1	故障检测率/%	57.6	0	45.8	39.0	93.4	76.1	84.2	93.6	99.4	99.0
	误报率/%	4.6	0	1.2	0.7	7.9	4.4	4.7	4.4	0	0
	DD/min	65	ND	42	46	38	113	0	0	8	10
2	故障检测率/%	48.4	0	10.6	36.4	68.9	72.7	28.9	76.1	99.4	86.6
	误报率/%	0.6	0	0.6	0.8	2.4	0.8	0	3.4	0.4	0
	DD/min	338	ND	447	294	161	142	80	8	10	74
3	故障检测率/%	38.2	0	18.6	20.6	83.4	76.1	69.3	63.5	99.6	99.2
	误报率/%	1	0	3.3	1.9	0	3.6	4.7	4.4	0.2	0
	DD/min	101	ND	93	ND	61	36	0	0	7	9
平均故障检测率/%		48.1	0	25.0	32.0	81.9	75.0	60.8	77.7	99.5	94.9

参考文献 30

1	Ge Z Q. Review on data-driven modeling and monitoring for plant-wide industrial processes[J]. Chemometrics and Intelligent Laboratory Systems, 2017, 171: 16-25.
2	Zhang Y, Bingham C, Garlick M, et al. Applied fault detection and diagnosis for industrial gas turbine systems[J]. International Journal of Automation and Computing, 2017, 14(4): 463-473.
3	Safaeipour H, Forouzanfar M, Casavola A. A survey and classification of incipient fault diagnosis approaches[J]. Journal of Process Control, 2021, 97: 1-16.
4	文成林, 吕菲亚, 包哲静, 等. 基于数据驱动的微小故障诊断方法综述[J]. 自动化学报, 2016, 42(9): 1285-1299.
	Wen C L, Lyu F Y, Bao Z J, et al. A review of data driven-based incipient fault diagnosis[J]. Acta Automatica Sinica, 2016, 42(9): 1285-1299.
5	李娟, 周东华, 司小胜, 等. 微小故障诊断方法综述[J]. 控制理论与应用, 2012, 29(12): 1517-1529.
	Li J, Zhou D H, Si X S, et al. Review of incipient fault diagnosis methods[J]. Control Theory & Applications, 2012, 29(12): 1517-1529.
6	Arunthavanathan R, Khan F, Ahmed S, et al. Fault detection and diagnosis in process system using artificial intelligence-based cognitive technique[J]. Computers & Chemical Engineering, 2020, 134: 106697.
7	Lou Z J, Wang Y Q, Si Y B, et al. A novel multivariate statistical process monitoring algorithm: orthonormal subspace analysis[J]. Automatica, 2022, 138: 110148.
8	Liu L, Liu J C, Wang H H, et al. A multivariate monitoring method based on kernel principal component analysis and dual control chart[J]. Journal of Process Control, 2023, 127: 102994.
9	Qin Y H, Yan Y Y, Ji H Q, et al. Recursive correlative statistical analysis method with sliding windows for incipient fault detection[J]. IEEE Transactions on Industrial Electronics, 2022, 69(4): 4185-4194.
10	衷路生, 解冬东. 基于鲁棒ICA-PCA的TE故障诊断[J]. 计算机应用研究, 2016, 33(10): 3026-3030.
	Zhong L S, Xie D D. Robust ICA-PCA based TE process monitoring and fault diagnosis[J]. Application Research of Computers, 2016, 33(10): 3026-3030.
11	Lei Y F, Jiang W L, Jiang A Q, et al. Fault diagnosis method for hydraulic directional valves integrating PCA and XGBoost[J]. Processes, 2019, 7(9): 589.
12	顾炳斌, 熊伟丽. 基于多块信息提取的PCA故障诊断方法[J]. 化工学报, 2019, 70(2): 736-749.
	Gu B B, Xiong W L. Fault diagnosis based on PCA method with multi-block information extraction[J]. CIESC Journal, 2019, 70(2): 736-749.
13	Benitez J, Henseler J, Castillo A, et al. How to perform and report an impactful analysis using partial least squares: guidelines for confirmatory and explanatory IS research[J]. Information & Management, 2020, 57(2): 103168.
14	Purwanto A. Partial least squares structural squation modeling (PLS-SEM) analysis for social and management research: a literature review[J]. Journal of Industrial Engineering & Management Research, 2021, 2(4): 114-123.
15	邓佳伟, 邓晓刚, 曹玉苹, 等. 基于加权统计局部核主元分析的非线性化工过程微小故障诊断方法[J]. 化工学报, 2019, 70(7): 2594-2605.
	Deng J W, Deng X G, Cao Y P, et al. Incipient fault diagnosis method of nonlinear chemical process based on weighted statistical local KPCA[J]. CIESC Journal, 2019, 70(7): 2594-2605.
16	Ge W S, Wang J, Zhou J L, et al. Incipient fault detection based on fault extraction and residual evaluation[J]. Industrial & Engineering Chemistry Research, 2015, 54(14): 3664-3677.
17	Zhao C H, Gao F R. A sparse dissimilarity analysis algorithm for incipient fault isolation with no priori fault information[J]. Control Engineering Practice, 2017, 65: 70-82.
18	Shang J, Chen M Y, Ji H Q, et al. Recursive transformed component statistical analysis for incipient fault detection[J]. Automatica, 2017, 80: 313-327.
19	Ji H Q, He X, Shang J, et al. Incipient fault detection with smoothing techniques in statistical process monitoring[J]. Control Engineering Practice, 2017, 62: 11-21.
20	张彩霞, 王子涵, 文成林, 等. 样本空间基于多级高维特征表示的微小故障诊断[J]. 电子学报, 2020, 48(8): 1647-1654.
	Zhang C X, Wang Z H, Wen C L, et al. Sample space based on multi-level high dimensional feature representation micro-fault diagnosis[J]. Acta Electronica Sinica, 2020, 48(8): 1647-1654.
21	鲍中新, 文成林, 马雪. 一种基于数据变化率的预处理及主元分析故障诊断方法[J]. 电子学报, 2021, 49(11): 2234-2240.
	Bao Z X, Wen C L, Ma X. Data preprocessing and PCA fault diagnosis method based on rate of change transformation[J]. Acta Electronica Sinica, 2021, 49(11): 2234-2240.
22	郭金玉, 王哲, 李元. 基于核熵独立成分分析的故障检测方法[J]. 化工学报, 2022, 73(8): 3647-3658.
	Guo J Y, Wang Z, Li Y. Fault detection method based on kernel entropy independent component analysis[J]. CIESC Journal, 2022, 73(8): 3647-3658.
23	张成, 潘立志, 李元. 基于加权统计特征KICA的故障检测与诊断方法[J]. 化工学报, 2022, 73(2): 827-837.
	Zhang C, Pan L Z, Li Y. Fault detection and diagnosis method based on weighted statistical feature KICA[J]. CIESC Journal, 2022, 73(2): 827-837.
24	Ge Z Q, Zhang M G, Song Z H. Nonlinear process monitoring based on linear subspace and Bayesian inference[J]. Journal of Process Control, 2010, 20(5): 676-688.
25	徐莹, 邓晓刚, 钟娜. 基于ICA混合模型的多工况过程故障诊断方法[J]. 化工学报, 2016, 67(9): 3793-3803.
	Xu Y, Deng X G, Zhong N. A fault diagnosis method for multimode processes based on ICA mixture models[J]. CIESC Journal, 2016, 67(9): 3793-3803.
26	Tong C D, Lan T, Shi X H. Ensemble modified independent component analysis for enhanced non-Gaussian process monitoring[J]. Control Engineering Practice, 2017, 58: 34-41.
27	Bach F. Information theory with kernel methods[J]. IEEE Transactions on Information Theory, 2023, 69(2): 752-775.
28	严少乐, 刘炜. 基于秩和比法的物流网络空间布局规划[J]. 物流技术, 2020, 39(9): 103-105, 111.
	Yan S L, Liu W. Logistics network spatial layout planning based on rank sum ratio[J]. Logistics Technology, 2020, 39(9): 103-105, 111.
29	Yue H H, Qin S J. Reconstruction-based fault identification using a combined index[J]. Industrial & Engineering Chemistry Research, 2001, 40(20): 4403-4414.
30	Pilario K E S, Cao Y. Canonical variate dissimilarity analysis for process incipient fault detection[J]. IEEE Transactions on Industrial Informatics, 2018, 14(12): 5308-5315.

[1]	曹跃, 余冲, 李智, 杨明磊. 工业数据驱动的加氢裂化装置多工况切换过渡状态检测[J]. 化工学报, 2023, 74(9): 3841-3854.
[2]	杨明辉, 刘晓月, 邓晓刚, 廖明燕, 侯春望. 基于加权概率CVDA的动态化工系统微小故障检测[J]. 化工学报, 2022, 73(9): 3963-3972.
[3]	刘立邦, 杨颂, 王志坚, 贺欣欣, 赵文磊, 刘守军, 杜文广, 米杰. 基于改进WOA-LSTM的焦炭质量预测[J]. 化工学报, 2022, 73(3): 1291-1299.
[4]	张成, 潘立志, 李元. 基于加权统计特征KICA的故障检测与诊断方法[J]. 化工学报, 2022, 73(2): 827-837.
[5]	李元, 杨东昇, 赵丽颖, 张成. 层次变分高斯混合模型与主多项式分析的故障检测策略[J]. 化工学报, 2021, 72(3): 1616-1626.
[6]	朱雄卓, 张瀚文, 杨春节. 基于高斯混合模型的MWPCA高炉异常监测算法[J]. 化工学报, 2021, 72(3): 1539-1548.
[7]	邓佳伟, 邓晓刚, 曹玉苹, 张晓玲. 基于加权统计局部核主元分析的非线性化工过程微小故障诊断方法[J]. 化工学报, 2019, 70(7): 2594-2605.
[8]	张晓晗, 汪平江, 顾祥柏, 徐圆, 贺彦林, 朱群雄. 基于主元提取的鲁棒极限学习机研究及其化工建模应用[J]. 化工学报, 2019, 70(2): 475-480.
[9]	耿志强, 景邵星, 白菊, 王仲凯, 朱群雄, 韩永明. 基于MWSPCA-CBR的智能预警方法研究及其在石化工业中的应用[J]. 化工学报, 2019, 70(2): 572-580.
[10]	顾炳斌, 熊伟丽. 基于多块信息提取的PCA故障诊断方法[J]. 化工学报, 2019, 70(2): 736-749.
[11]	翟坤, 杜文霞, 吕锋, 辛涛, 句希源. 一种改进的动态核主元分析故障检测方法[J]. 化工学报, 2019, 70(2): 716-722.
[12]	高学金, 黄梦丹, 齐咏生, 王普. PDPSO优化多阶段AR-PCA间歇过程监测方法[J]. 化工学报, 2018, 69(9): 3914-3923.
[13]	董顺, 李益国, 孙栓柱, 刘西陲, 沈炯. 基于状态空间主成分分析网络的故障检测方法[J]. 化工学报, 2018, 69(8): 3528-3536.
[14]	邓晓刚, 邓佳伟, 曹玉苹, 王磊. 基于双层局部KPCA的非线性过程微小故障检测方法[J]. 化工学报, 2018, 69(7): 3092-3100.
[15]	冯立伟, 张成, 李元, 谢彦红. 基于双近邻标准化和PCA的多阶段过程故障检测[J]. 化工学报, 2018, 69(7): 3159-3166.