面向随钻边缘计算的井下钻井工况轻量化智能识别方法

doi:10.3969/j.issn.2096-1693.2026.02.016

石油科学通报 ›› 2026, Vol. 11 ›› Issue (3): 850-766. doi: 10.3969/j.issn.2096-1693.2026.02.016

面向随钻边缘计算的井下钻井工况轻量化智能识别方法

詹家豪¹(), 李军¹^,²^,^*(), 柳贡慧¹^,³, 杨宏伟¹, 王超⁴, 王彪¹

¹ 中国石油大学(北京)石油工程学院，北京 102249
² 中国石油大学(北京)克拉玛依校区石油学院，克拉玛依 834000
³ 北京工业大学机械与能源工程学院，北京 100124
⁴ 长江大学机械工程学院，荆州 434023

收稿日期:2025-11-25 修回日期:2026-01-19 出版日期:2026-06-15 发布日期:2026-06-30
通讯作者: *李军(1971年—)，教授，博导，主要研究方向为欠平衡/控压钻井，井下工况人工智能识别、井筒完整性，射孔完井，页岩气开发，岩石力学及其应用，lijun17792692628@163.com。
作者简介:詹家豪(1999年—)，在读博士研究生，主要研究方向为油气工程信息化与智能化技术，zjh37730904@163.com。
基金资助:
国家重点研发计划项目“陆上超深油气井井喷防控关键技术装备及示范应用”(2023YFC3009200);国家自然科学基金重大科研仪器研制项目“钻井复杂工况井下实时智能识别系统研制”(52227804);国家自然科学基金青年科学基金项目“深井气侵井下原位实时识别与定量解释方法研究”(52304001);国家自然科学基金面上项目“超深复杂地层溢流智能识别与关井—压井一体化调控方法”(52474018)

A lightweight intelligent identification method for downhole drilling conditions based on MWD-oriented edge computing

ZHAN Jiahao¹(), LI Jun¹^,²^,^*(), LIU Gonghui¹^,³, YANG Hongwei¹, WANG Chao⁴, WANG Biao¹

¹ College of Petroleum Engineering, China University of Petroleum, Beijing 102249, China
² College of Petroleum Engineering, China University of Petroleum-Beijing at Karamay, Karamay 834000, China
³ College of Mechanical and Energy Engineering, Beijing University of Technology, Beijing 100124, China
⁴ School of Mechanical Engineering, Yangtze University, Jingzhou 434023, China

Received:2025-11-25 Revised:2026-01-19 Online:2026-06-15 Published:2026-06-30
Contact: *lijun17792692628@163.com

摘要/Abstract

摘要：

深层钻井过程中工况复杂多变，基于地面参数的识别方法存在传播滞后与误判率高的问题，而井下由于计算资源受限难以支撑复杂算法的直接部署。本文提出一种数据驱动的改进型层次分析法(Data-driven Modified Analytic Hierarchy Process，DM-AHP)：以SHAP(SHapley Additive exPlanation)值替代专家评分构建AHP判断矩阵，利用SHAP值的数学性质(效率性、对称性、线性性与零贡献性)从理论上保证权重分配的客观性与乘法一致性。基于"地面训练—井下识别"的分布式架构，地面端完成模型训练、特征解释与判断矩阵生成，井下端仅部署160个参数的轻量化矩阵，并结合基于箱线图的动态阈值机制，实现实时工况识别与自适应更新。基于11口井实测数据验证：模型在7类正常工况识别中准确率达95.5%，F1=0.954；异常状态检测F1=0.931；内存占用48 KB，单次推理耗时42 ms，满足井下实时约束。跨井泛化测试表明，双重动态更新策略(归一化参数更新与阈值自适应调整)将F1由0.864提升至0.954。相较于传统机器学习模型10⁴~10⁶量级的参数规模，本方法在参数压缩逾99%的条件下，识别性能仅下降约1.7%，可直接嵌入随钻测量系统，为深层钻井智能工况监测提供兼具可解释性、轻量化与高精度的可实施技术方案。

关键词: 钻井工况识别, 数据驱动, 层次分析法, SHAP值, 边缘计算, 轻量化模型, 随钻识别

Abstract:

Deep drilling operations involve complex and variable working conditions, where surface-based identification methods suffer from signal propagation delays and high misclassification rates, while limited downhole computing resources preclude the direct deployment of complex algorithms. This paper proposes a Data-driven Modified Analytic Hierarchy Process (DM-AHP) that replaces expert scoring with SHapley Additive exPlanation (SHAP) values for constructing AHP judgment matrices. The mathematical properties of SHAP values—efficiency, symmetry, linearity, and the null player property—theoretically guarantee the objectivity and multiplicative consistency of weight allocation. A “surface training-downhole identification” distributed architecture is established: model training, feature interpretation, and judgment matrix generation are performed on the surface, while only a lightweight 160-parameter matrix is deployed downhole, coupled with a boxplot-based dynamic threshold mechanism for real-time condition identification and adaptive updating. Validation on field data from 11 wells demonstrates that the model achieves 95.5% accuracy and an F1-score of 0.954 in classifying seven normal drilling conditions, with an anomaly detection F1-score of 0.931. Memory consumption is 48 KB and single inference latency is 42 ms, satisfying downhole real-time constraints. Cross-well generalization tests show that the dual dynamic update strategy—combining normalization parameter updates with adaptive threshold adjustment—improves the F1-score from 0.864 to 0.954. Compared with conventional machine learning models with parameter counts on the order of 10⁴~10⁶, the proposed method reduces parameters by over 99% with only a 1.7% decrease in recognition performance. The method can be directly embedded into measurement-while-drilling systems, offering a practically deployable solution that simultaneously achieves interpretability, lightweight deployment, and high accuracy for intelligent condition monitoring in deep drilling operations.

Key words: drilling condition identification, data-driven, analytic hierarchy process, SHAP values, edge computing, lightweight model, real-time drilling identification

中图分类号:

TE21
TP18

詹家豪, 李军, 柳贡慧, 杨宏伟, 王超, 王彪. 面向随钻边缘计算的井下钻井工况轻量化智能识别方法[J]. 石油科学通报, 2026, 11(3): 850-766.

ZHAN Jiahao, LI Jun, LIU Gonghui, YANG Hongwei, WANG Chao, WANG Biao. A lightweight intelligent identification method for downhole drilling conditions based on MWD-oriented edge computing[J]. Petroleum Science Bulletin, 2026, 11(3): 850-766.

https://sykxtb.cup.edu.cn/CN/Y2026/V11/I3/850

图/表 19

图1 钻井工况识别的AHP层次结构

Fig. 1 AHP hierarchical structure for drilling condition identification

表1 数据驱动模型特性对比分析

Table 1 Comparative analysis of data-driven model characteristics

模型类型	动态建模能力	可解释性	特征交互捕捉	适用场景
随机森林	弱	强	弱	特征重要性分析
SVM	弱	中	弱	高维数据分析
CNN	中	弱	强	复杂特征关联

表2 随钻测量系统数据指标

Table 2 Data indicators of measurement while drilling system

参数类别	代表性参数	测量频率/Hz
流体压力参数	环空压力、钻柱内压力	20
钻头载荷参数	钻压、扭矩	20
钻头运动参数	转速	1000
管柱振动参数	X/Y/Z振动	1000

图2 井下测量的低通滤波效果

Fig. 2 Low-pass filtering effects on downhole measurements

表3 井下正常工况人工标注逻辑

Table 3 Manual annotation logic for normal downhole drilling conditions

参数	静止	起下钻	开停泵	划眼	循环	滑动钻进	复合钻进
钻柱内外压差	≈0	≈0	↑	>0	>0	>0	>0
转速	≈0	≈0	↑	>0	>0	<螺杆转速上限	>螺杆转速上限
钻压	≈0	≈0	≈0	≈0	≈0	>0	>0
扭矩	≈0	≈0	≈0	>0	≈0	>0	>0
X/Y/Z振动	无	弱	中	强	中	强	强

图3 7类正常工况统计比例

Fig. 3 Statistical proportion of seven normal drilling conditions

图4 地面训练—井下识别的分布式架构流程

Fig. 4 Distributed architecture workflow of surface training and downhole identification

图5 循环工况SHAP值构建AHP判断矩阵示例

Fig. 5 Example of constructing AHP judgment matrix using SHAP values for circulation condition

图6 滑动时间窗口数据划分示意图

Fig. 6 Schematic diagram of sliding time window data segmentation

图7 井下端实时工况识别数据流

Fig. 7 Downhole real-time condition identification workflow

图8 基于得分的工况判断流程

Fig. 8 Score-based condition determination workflow

表4 不同算法性能与资源消耗对比

Table 4 Performance and resource consumption comparison of different algorithms

方法	训练时间/min	参数数量	运行峰值内存/KB	单次计算时间/ms	井下应用可行性	准确率/%	召回率/%	F1得分
SVM	15.2	3.09×10⁶	12 974	>1000	不可行	94.3	94.6	0.945
CNN	45.8	5.11×10⁴	467	>1000	不可行	95.2	95.5	0.953
RF	8.7	3.36×10⁶	11 039	>1000	不可行	94.8	94.2	0.950
Expert-AHP	-	160	48	<50	可行	72.4	75.9	0.741
SVM-AHP	18.6	160	48	<50	可行	90.1	90.2	0.901
CNN-AHP	54.0	160	48	<50	可行	92.6	93.1	0.928
RF-AHP	10.8	160	48	<50	可行	93.5	93.3	0.934

图9 RF-AHP模型7类工况识别性能评估

Fig. 9 Performance evaluation of RF-AHP model for 7 condition types identification

图10 不同时间窗口长度对RF-AHP模型识别性能的影响

Fig. 10 Impact of different time window lengths on RF-AHP model identification performance

图11 基于30 s时间窗口的特征权重系数矩阵热力图

Fig. 11 Heatmap of feature weight coefficient matrix based on 30 s time window

图12 Well-11连续38 h实时工况识别时序结果

Fig. 12 Time Series Results of Real-Time Condition Identification for Well-11 over 38 h

图13 RF-AHP模型动态更新策略运行效果

Fig. 13 Performance comparison of dynamic update strategy for RF-AHP model

表5 各算法井间泛化性能对比

Table 5 Cross-well generalization performance comparison of different algorithms

模型	正常工况准确率/%	正常工况召回率/%	正常工况F1	异常状态检测F1
SVM	79.3	80.8	0.800	-
CNN	82.9	83.5	0.832	-
RF	82.6	84.9	0.837	-
RF-AHP(静态)	86.7	86.2	0.864	0.716
RF-AHP(动态)	95.5	95.4	0.954	0.931

图14 工况识别结果混淆矩阵

Fig. 14 Confusion matrix of condition identification results

参考文献 31

[1]	高德利, 黄文君. 超深井工程理论与技术若干研究进展及发展建议[J]. 石油钻探技术, 2024, 52(2): 1-11, F0003.
	[GAO D L, HUANG W J. Research and development suggestions on theory and techniques in ultra-deep well engineering[J]. Petroleum Drilling Techniques, 2024, 52(2): 1-11, F0003.]
[2]	王建龙, 王越支, 邱卫红, 等. 基于大数据与融合模型的钻井智能辅助决策系统[J]. 石油钻探技术, 2024, 52(5): 105-116.
	[WANG J L, WANG Y Z, QIU W H, et al. Drilling intelligent decision support system based on big data and fusion model[J]. Petroleum Drilling Techniques, 2024, 52(5): 105-116.]
[3]	王正, 宋先知, 李洪松, 等. 基于1dCNN-BiGRU和注意力机制的钻井工况智能识别方法[J]. 石油科学通报, 2025, 10(5): 926-940.
	[WANG Z, SONG X Z, LI H S, et al. Intelligent recognition method for drilling conditions based on 1dCNN-BiGRU and attention mechanism[J]. Petroleum Science Bulletin, 2025, 10(5): 926-940.]
[4]	YIN Q S, YANG J, LIU S J, et al. Intelligent method of identifying drilling risk in complex formations based on drilled wells data[C]// SPE Intelligent Oil and Gas Symposium, Abu Dhabi, UAE, 2017: SPE-187472-MS.
[5]	LI Y P, CAO W H, HU W K, et al. Identification of downhole conditions in geological drilling processes based on quantitative trends and expert rules[J]. Neural Computing and Applications, 2023, 35(17): 12297-12306. doi: 10.1007/s00521-021-05759-4
[6]	YOUCEFI M R, BOUKREDERA F S, GHALEM K, et al. Development of an expert-informed rig state classifier using naive Bayes algorithm for invisible loss time measurement[J]. Applied Intelligence, 2024, 54(17): 7659-7673. doi: 10.1007/s10489-024-05560-5
[7]	祝兆鹏, 张瑞, 宋先知, 等. 数据拓展和增量更新的井底压力实时预测方法[J]. 石油机械, 2023, 51(6): 1-10.
	[ZHU Z P, ZHANG R, SONG X Z, et al. Real-time prediction of bottomhole pressure based on data expansion and incremental update[J]. China Petroleum Machinery, 2023, 51(6): 1-10.]
[8]	YANG X Y, ZHANG Y L, ZHOU D T, et al. Drilling conditions classification based on improved stacking ensemble learning[J]. Energies, 2023, 16(15): 5747. doi: 10.3390/en16155747 URL
[9]	LIU Z H, SONG X Z, YE S L, et al. Intelligent identification workflow of drilling conditions combining deep learning and drilling knowledge[C]// ASME 2024 43rd International Conference on Ocean, Offshore and Arctic Engineering, Singapore, Singapore, 2024
[10]	GHASEMLOONIA A, RIDEOUT D G, BUTT S D. Analysis of multi-mode nonlinear coupled axial-transverse drillstring vibration in vibration assisted rotary drilling[J]. Journal of Petroleum Science and Engineering, 2014, 116: 36-49. doi: 10.1016/j.petrol.2014.02.014 URL
[11]	NAYEEM A A, VENKATESAN R, KHAN F. Monitoring of down-hole parameters for early kick detection[J]. Journal of Loss Prevention in the Process Industries, 2016, 40: 43-54. doi: 10.1016/j.jlp.2015.11.025 URL
[12]	WSPANIALY A, KYAW M. Surface drilling parameters and drilling optimization techniques: Are they useful tools in gas hydrate detection[J]. Energies, 2022, 15(13): 4635. doi: 10.3390/en15134635 URL
[13]	苏义脑, 窦修荣, 高文凯, 等. 油气井随钻测量技术发展思考与展望[J]. 石油科学通报, 2023, 8(5): 535-554.
	[SU Y N, DOU X R, GAO W K, et al. Discussion and prospects of the development on measurement while drilling technology in oil and gas wells[J]. Petroleum Science Bulletin, 2023, 8(5): 535-554.]
[14]	MWACHAKA S M, WU A P, FU Q Q. A review of mud pulse telemetry signal impairments modeling and suppression methods[J]. Journal of Petroleum Exploration and Production Technology, 2019, 9(1): 779-792. doi: 10.1007/s13202-018-0483-y
[15]	WANG C, LIU G H, YANG Z R, 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: 103485. doi: 10.1016/j.jngse.2020.103485 URL
[16]	SHUVO M M H, ISLAM S K, CHENG J L, et al. Efficient acceleration of deep learning inference on resource-constrained edge devices: A review[J]. Proceedings of the IEEE, 2023, 111(1): 42-91. doi: 10.1109/JPROC.2022.3226481 URL
[17]	YANG W J, WANG Y Q, ZHANG X H, et al. MSS-YOLO: Multi-scale edge-enhanced lightweight network for personnel detection and location in coal mines[J]. Applied Sciences, 2025, 15(6): 3238. doi: 10.3390/app15063238 URL
[18]	LING J R, FU Z B, YUAN X P, et al. Development of a deep learning-based foreign object detection algorithm for coal mine conveyor belts[J]. Scientific Reports, 2025, 15: 42291. doi: 10.1038/s41598-025-22636-5
[19]	李瑷昕. 基于区域统计数据的水平井钻井难度评价体系及方法研究[D]. 西安: 西安石油大学, 2023.
	[LI A X. Research on evaluation and method of horizontal well drilling difficulty based on regional statistical data[D]. Xi’an: Xi’an Shiyou University, 2023.]
[20]	WIĘCKOWSKI J, KIZIELEWICZ B, WĄTRÓBSKI J, et al. A new approach for handling uncertainty of expert judgments in complex decision problems[J]. IEEE Access, 2024, 12: 142026-142046. doi: 10.1109/ACCESS.2024.3445265 URL
[21]	GUPTA R, BHUI U K. Transforming oil well drilling: Prediction of real-time rate of penetration with novel machine learning approach in varied lithological formations[J]. Journal of Energy Resources Technology, Part B: Subsurface Energy and Carbon Capture, 2025, 1: 011003. doi: 10.1115/1.4066015 URL
[22]	ZHANG J Y, MA X L, ZHANG J L, et al. Insights into geospatial heterogeneity of landslide susceptibility based on the SHAP-XGBoost model[J]. Journal of Environmental Management, 2023, 332: 117357. doi: 10.1016/j.jenvman.2023.117357 URL
[23]	LI M, SUN H Y, HUANG Y J, et al. Shapley value: From cooperative game to explainable artificial intelligence[J]. Autonomous Intelligent Systems, 2024, 4(1): 2. doi: 10.1007/s43684-023-00060-8
[24]	陈君青, 杨晓斌, 张潇, 等. 页岩力学性质研究中机器学习的应用: 现状、挑战与展望[J]. 石油科学通报, 2025, 10(5): 849-877.
	[CHEN J Q, YANG X B, ZHANG X, et al. Application of machine learning in the study of shale mechanical properties: Current situation, challenges and prospects[J]. Petroleum Science Bulletin, 2025, 10(5): 849-877.]
[25]	刘子豪, 宋先知, 朱硕, 等. 基于VC-SVM与粒子群算法的卡钻智能预测方法[J]. 石油机械, 2024, 52(10): 1-11.
	[LIU Z H, SONG X Z, ZHU S, et al. Intelligent prediction method of pipe sticking based on VC-SVM and particle swarm optimization[J]. China Petroleum Machinery, 2024, 52(10): 1-11.]
[26]	孟翰, 张振鑫, 韩雪银, 等. 融合钻头磨损与预训练机制的机械钻速预测方法研究[J]. 石油科学通报, 2025, 10(5): 1030-1046.
	[MENG H, ZHANG Z X, HAN X Y, et al. Research on rate of penetration prediction method integrating bit wear and pretraining mechanism[J]. Petroleum Science Bulletin, 2025, 10(5): 1030-1046.]
[27]	SHAMSI E, ALI AHMADI-PAJOUH M, TOWHIDKHAH F, et al. Time-reassigned adaptive Fourier synchrosqueezing transform with globally optimal window length[J]. Biomedical Signal Processing and Control, 2024, 97: 106692. doi: 10.1016/j.bspc.2024.106692 URL
[28]	张来斌, 胡瑾秋, 肖尚蕊, 等. 深部地下空间储能安全与应急保障技术现状与发展趋势[J]. 石油科学通报, 2024, 9(3): 434-448.
	[ZHANG L B, HU J Q, XIAO S R, et al. Current status and development trend of safety and emergency support technology for energy storage in deep underground spaces[J]. Petroleum Science Bulletin, 2024, 9(3): 434-448.]
[29]	HU Y, ZHAO X W, LI Q. Research on adaptive prediction model of rate of penetration under dynamic formation conditions[J]. Engineering Applications of Artificial Intelligence, 2024, 133: 108281. doi: 10.1016/j.engappai.2024.108281 URL
[30]	BERTINI J R, LAVI B. Enhancing rate of penetration prediction in drilling operations: A data stream framework approach[J]. Engineering Applications of Artificial Intelligence, 2025, 143: 110034. doi: 10.1016/j.engappai.2025.110034 URL
[31]	ALI A, GRAVINO C. An empirical comparison of validation methods for software prediction models[J]. Journal of Software: Evolution and Process, 2021, 33(8): e2367. doi: 10.1002/smr.v33.8 URL

选择文件类型/文献管理软件名称

选择包含的内容

面向随钻边缘计算的井下钻井工况轻量化智能识别方法

A lightweight intelligent identification method for downhole drilling conditions based on MWD-oriented edge computing

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 19

参考文献 31

相关文章 4

编辑推荐

Metrics

本文评价

[1]	詹家豪, 李军, 柳贡慧, 杨宏伟, 王超, 王彪. 基于随钻实时数据的井下托压多级智能识别方法[J]. 石油科学通报, 2026, 11(2): 518-532.
[2]	胡益, 侯磊, 于巧燕, 喻鹏飞, 魏平洋, 杨牟青云, 蒋璐朦. 基于排序节点法和层次分析法的输气管道模糊贝叶斯网络风险分析[J]. 石油科学通报, 2026, 11(2): 614-628.
[3]	魏然然, 唐帅帅, 郑洪龙, 侯磊, 刘玥奇, 周子栋, 程遹滔. 复杂结构天然气管网系统递阶式可靠性分配方法[J]. 石油科学通报, 2025, 10(6): 1374-1388.
[4]	周军；彭井宏；孙建华；肖瑶；梁光川. 基于模糊综合评价法的气藏型储气库注采方案优选研究[J]. , 2021, 6(3): 494-504.

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

面向随钻边缘计算的井下钻井工况轻量化智能识别方法

A lightweight intelligent identification method for downhole drilling conditions based on MWD-oriented edge computing

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 19

参考文献 31

相关文章 4

编辑推荐

Metrics

本文评价