基于BP-LSTM双输入网络的大钩载荷与转盘扭矩预测  

Prediction of hook load and rotary drive torque during well-drilling using a BP-LSTM network

作　　者：宋先知[1,2] 朱硕 李根生[1,2] 曾义金[3] 郭慧娟 胡志坚 SONG Xianzhi;ZHU Shuo;LI Gensheng;ZENG Yijin;GUO Huijuan;HU Zhijian(School of Petroleum Engineering, China University of Petroleum (Beijing), Beijing 102249, China;State Key Laboratory of Petroleum Resources and Prospecting, China University of Petroleum (Beijing), Beijing 102249, China;SINOPEC Research Institute of Petroleum Engineering, Beijing 100101, China;CNPC Engineering Technology R & D Company Limited, Beijing 102206, China)

机构地区：[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)油气资源与探测国家重点实验室,北京102249 [3]中国石化石油工程技术研究院,北京100101 [4]中国石油集团工程技术研究院有限公司,北京102206

出　　处：《中国石油大学学报（自然科学版）》2022年第2期76-84,共9页Journal of China University of Petroleum(Edition of Natural Science)

基　　金：中国石油天然气集团有限公司-中国石油大学(北京)战略合作科技专项(ZLZX2020-03);国家重点研发计划(2019YFA0708300)。

摘　　要：考虑影响钩载、扭矩的因素复杂多样及钻井过程的时序性特点,优选BP神经网络和长短期记忆神经网络,设计双输入网络架构,建立大钩载荷与转盘扭矩智能预测模型。该模型同时考虑影响钩载、扭矩的多种复杂参数以及钩载、扭矩等时序数据随时间变化的趋势和前后关联,通过时序性数据和非时序性数据共同预测大钩载荷与转盘扭矩。利用国内某油田钻井现场数据进行大钩载荷与转盘扭矩的预测,均方根误差分别为39.05 kN和1.6274 kN·m,平均相对误差分别为1.202%和9.038%。In this paper,BP neural network and long&short-term memory neural network were selected along with a double-input network architecture to establish an intelligent prediction model of the hook load and rotary drive torque.In the modeling,a variety of complex parameters affecting the torque and load,and their dynamic variations with time were considered simultaneously,with the torque and drag being predicted using both sequential data and non-sequential data.A case study using a real oilfield drilling site data indicates that,the hook load and rotary torque can be accurately predicted,with root mean square errors of 39.05 kN and 1.6274 kN·m,and average relative errors of 1.202%and 9.038%,respectively.

关 键 词：大钩载荷 转盘扭矩 BP神经网络 长短期记忆神经网络 人工智能 

分 类 号：TE21[石油与天然气工程—油气井工程]