A multi-class intelligent identification model for kick risk

doi:10.16265/j.cnki.issn1003-3033.2026.01.0911

Abstract

Abstract:

In order to improve the identification accuracy of kick risk during drilling, a multi-category kick risk intelligent identification model was proposed by integrating feature engineering and machine learning techniques. Firstly, a wavelet transform was employed to achieve noise suppression based on field-measured kick data. Secondly, the dynamic variation trends of key parameters were extracted using smooth spline functions, and the abnormal fluctuation behaviors of kick-related characteristic parameters were analyzed. Based on this, a three-level risk classification criterion (low, medium, and high) was established, and kick risks were labeled according to the variation features of drilling data. Then, the sparrow search algorithm (SSA) was introduced to optimize the extreme learning machine (ELM), and a multi-classification kick risk intelligent identification model based on IELM was constructed. Finally, the performance of the model was validated through training, tuning, and testing on the constructed risk dataset. The results show that the IELM model outperforms the original ELM and back-propagation (BP) neural network model in terms of classification accuracy and discrimination stability, and is capable of identifying different levels of kick risks more accurately and efficiently.

Key words: overflow risk, identification model, improved extreme learning machine (IELM), neural network, characteristic parameters

CLC Number:

X937

WU Shengnan, ZHANG Laibin, HU Yiming, CUI Rong, LIU Shujie, YIN Zhiming. A multi-class intelligent identification model for kick risk[J]. China Safety Science Journal, 2026, 36(1): 72-80.

Figures/Tables 22

Fig.1

Fig.2

Table 1

Table 2

Fig.3

Table 3

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Fig.9

Fig.10

Fig.11

Fig.12

Table 4

Fig.13

Fig.14

Table 5

Fig.15

Fig.16

Table 6

References 21

[1]	XU Chengyuan, YAN Xiaopeng, KANG Yili, et al. Structural failure mechanism and strengthening method of fracture plugging zone for lost circulation control in deep naturally fractured reservoirs[J]. Petroleum Exploration and Development, 2020, 47(2): 430-440. doi: 10.1016/S1876-3804(20)60060-X
[2]	李玉飞, 张博, 孙伟峰. 基于SVM和D-S证据理论的早期溢流智能识别方法[J]. 钻采工艺, 2020, 43(5):27-30,6. doi: 10.3969/J. ISSN.1006-768X.2020.05.08
	LI Yufei, ZHANG Bo, SUN Weifeng. Research on intelligent early kick identification method based on SVM and D-S evidence theory[J]. Drilling & Production Technology, 2020, 43(5): 27-30,6.
[3]	邴磊. 基于Attention-GRU算法早期溢流识别预警方法[J]. 海洋石油, 2025, 45(2):76-82.
	BING Lei. Early overflow identification based on Attention-GRU algorithm early warning method[J]. Offshore Oil, 2025, 45(2):76-82.
[4]	张禾, 池紫欣. 基于BSMOTE-SVM算法的溢流风险评价[J]. 控制工程, 2023, 30(12): 2173-2178.
	ZHANG He, CHI Zixin. Overflow risk evaluation based on BSMOTE-SVM algorithm[J]. Control Engineering of China, 2023, 30(12): 2173-2178.
[5]	YANG Jin, SUN Ting, ZHAO Ying, et al. Advanced real-time gas kick detection using machine learning technology[C]. Proceedings of the 29^th International Ocean and Polar Engineering Conference, 2019:1404-1409.
[6]	陈青, 黄志强, 孔祥伟, 等. 基于多录井参数特征同步的溢流事故监测研究[J]. 应用数学和力学, 2025, 46(2): 241-53.
	CHEN Qing, HUANG Zhiqiang, KONG Xiangwei, et al. Study on overflow accident monitoring based onsynchronous features of multiple well logging parameters[J]. Applied Mathematics and Mechanics, 2025, 46(2): 241-53.
[7]	ZHANG Dezhi, SUN Weifeng, DAI Yongshou, et al. A hierarchical early kick detection method using a cascaded GRU network[J]. Geoenergy Science and Engineering, 2023, 222: DOI:10.1016/j.geoen.2022.211390.
[8]	史肖燕, 周英操, 赵莉萍, 等. 基于随机森林的溢漏实时判断方法研究[J]. 钻采工艺, 2020, 43(1): 9-12,7. doi: 10.3969/J. ISSN.1006-768X.2020.01.03
	SHI Xiaoyan, ZHOU Yingcao, ZHAO Liping, et al. Losses in real time on basis of random forest method[J]. Drilling & Production Technology, 2020, 43(1): 9-12,7.
[9]	WANG Chao, LIU Gonghui, YANG Zhirong, et al. Downhole working conditions analysis and drilling complications detection method based on deep learning[J]. Journal of Natural Gas Science and Engineering, 2020, 81: DOI:10.1016/j.jngse.2020.103485.
[10]	ZHANG Zhi, SUN Baojiang, WANG Zzhiyuan, et al. Early monitoring method of downhole accident driven by physics based model and data driven methods coupling[J]. Geoenergy Science and Engineering, 2023, 221:DOI:10.1016/j.jngse.2020.103485.
[11]	LIU Wei, FU Jiasheng, DENG Song, et al. Overflow identification and early warning of managed pressure drilling based on series fusion data-driven mode[J]. Processes, 2024, 12: DOI:10.3390/pr12071436.
[12]	陈贵波, 刘东升, 李加卫, 等. 基于时间和通道融合的多变量时间序列异常检测[J/OL]. 计算机应用与软件:1-13.[2025-09-01]. https://link.cnki.net/urlid/31.1260.TP.20250919.1739.004.
	CHEN Guibo, LIU Dongsheng, LI Jiawei, et al. Multivariate time series anomaly detection based on time and channel fusion[J/OL]. Computer Applications and Software:1-13.[2025-09-01]. https://link.cnki.net/urlid/31.1260.TP.20250919.1739.004.
[13]	WANG Shuhua, CHEN Zan, CHEN Shengnan. Applicability of deep neural networks on production forecasting in Bakken shale reservoirs[J]. Journal of Petroleum Science and Engineering, 2019, 179: 112-125. doi: 10.1016/j.petrol.2019.04.016
[14]	杨向前, 张苹茹, 武胜男, 等. 基于数据模型协作的海上钻井溢流早期预测预警[J]. 中国安全科学学报, 2024, 34(4): 93-100. doi: 10.16265/j.cnki.issn1003-3033.2024.04.1390
	YANG Xiangqian, ZHANG Pingru, WU Shengnan, et al. Early prediction and waring of ofshore drilling overlow based ondata model collaboration[J]. China Safety Science Joural, 2024, 34(4):93-100.
[15]	葛亮, 滕怡, 肖国清, 等. 基于井下环空参数的溢流智能预警技术研究[J]. 西南石油大学学报:自然科学版, 2023, 45(2): 126-134.
	GE Liang, TENG Yi, XIAO Guoqing, et al. Research on overlow intelligent warning teclmology based on downhole annulus parameters[J]. Joumnal of Southwest Petroleum university:Science & Technology Edition, 2023, 45(2):126-134.
[16]	李开荣, 陈俊男, 段丽娟, 等. 钻井液池体积精准监测装置的研制与应用[J]. 录井工程, 2022, 33(4):92-96. doi: 10.3969/j.issn.1672-9803.2022.04.015
	LI Kairong, CHEN Junnan, DUAN Lijuan, et al. Development and application of accurate monitoring device for drilling fluid pit volume[J]. Mud Logging Engineering, 2022, 33(4):92-96. doi: 10.3969/j.issn.1672-9803.2022.04.015
[17]	张继德, 张永刚, 韩国生, 等. 钻井液录井参数在钻井工程异常预报中的应用[J]. 录井工程, 2010, 21(3): 39-44,76-77.
	ZHANG Jide, ZHANG Yonggang, HAN Guosheng, et al. The application of drilling fluid logging parameter in the abnormal prediction of drilling engineering[J]. Mud Logging Engineering, 2010, 21(3): 39-44,76-77.
[18]	OSAROGIAGBON A, MUOJEKE S, VENKATESAN R, et al. A new methodology for kick detection during petroleum drilling using long short-term memory recurrent neural network[J]. Process Safety and Environmental Protection, 2020, 142: 126-137. doi: 10.1016/j.psep.2020.05.046
[19]	SUN Weifeng, LI Weihua, ZHANG Dezhi, et al. Lost circulation monitoring using bi-directional LSTM and data augmentation[J]. Geoenergy Science and Engineering, 2023, 225: DOI:10.1016/j.geoen.2023.211660.
[20]	WU Shengnan, HU Yiming, ZHANG Laibin, et al. Intelligent risk identification for drilling lost circulation incidents using data-driven machine learning[J]. Reliability Engineering & System Safety, 2024, 252: DOI:10.1016/j.ress.2024.110407.
[21]	POWERS D. Evaluation: from precision, recall and F-measure to ROC, informedness, markedness & correlation[J]. Journal Machine Learning Technologies, 2011, 2: 37-63.

表征参数	参数异常形式	重要级别
气测值	数据呈现快速上升趋势、数据大于一定阈值、离群异常数据	Ⅰ-核心参数
活动泥浆池体积	数据呈现上升趋势、离群异常数据	Ⅱ-典型参数
钻井液相对出口流量	数据呈现上升趋势、离群异常数据	Ⅲ-重要参数
立管压力	数据呈现下降趋势、离群异常数据	Ⅳ-辅助参数

溢流分级	划分规则	风险标签
低溢流风险	不多于1个非核心参数出现异常	1
中溢流风险	核心参数出现异常	2
中溢流风险	核心参数与典型参数不同时异常时2个参数出现异常	2
高溢流风险	核心参数与典型参数同时出现异常	3
高溢流风险	3个及以上参数出现异常	3

数据序号	时间范围	井筒深度/m	钻头深度/m	工况
1—714	8:54:15—9:53:40	3 995.29~4 002.43	3 995.29~4 002.43	纯钻进
715—868	9:53:45—10:06:30	4 002.43	3 992.93~4 002.43	划眼
869—976	10:06:35—10:15:30	4 002.43~4 003.45	4 002.43~4 003.45	纯钻进
977—1 003	10:15:35—10:17:45	4 003.45	4 002.78~4 003.45	划眼
1 004—2 678	10:17:50—12:37:20	4 003.45~4 023.55	4 003.45~4 023.55	纯钻进
2 679—3 825	12:37:25—14:12:55	4 023.55	3 994.31~4 023.55	划眼(其间溢流)
3 826—5 839	14:13:00—17:00:55	4 023.55	4 020.93~4 021.41	循环(其间溢流)

数据集	准确率	IELM	ELM	BP	SVM
数据训练集A₁	低风险	99.64	81.58	92.57	98.49
	中风险	96.89	92.81	92.67	96.11
	高风险	87.63	0	24.73	89.49
	整体	97.16	79.81	86.44	96.48
数据测试集A₂	低风险	99.01	82.79	92.10	95.03
	中风险	96.03	92.53	92.30	91.32
	高风险	82.5	0	23.75	72.67
	整体	96.01	80.14	85.96	91.10
	F₁分数	93.01	—	71.47	87.03