摘要
钻井过程中发生钻井液漏失时,现有的井漏智能监测方法,难以获取长时数据序列特征,无法实现对微量漏失的及时监测和预警,进而容易导致更为严重的漏失发生。为此,提出了一种结合扩张因果卷积网络(Dilated and Causal Convolution,DCC)特征映射能力和长短期记忆网络(Long Short-Term Memory,LSTM)时序特征提取能力的DCC-LSTM钻井液微量漏失智能监测方法,弥补长短期记忆网络对于长期记忆衰减的不足,实现了对钻井液微量漏失的准确监测和预测。研究结果表明:①DCC-LSTM井漏智能监测模型利用扩张因果卷积网络提取监测参数的长时特征,并将其映射为短序列表示,利用长短期记忆网络处理特征短序列获取监测数据的长时变化趋势,实现了微量漏失的准确监测;②扩张因果卷积网络层数确定方法可以获得最佳网络层数,得到的DCC网络结构使LSTM对长时序列趋势信息的遗忘减少24%;③与其他井漏监测方法相比,DCC-LSTM网络能够准确监测早期微量漏失,井漏预警时间最长可提前26 min,监测准确率由96.9%提升至99.4%,漏报率由6.4%降低为1.1%。结论认为,该方法能够获取监测参数的长时趋势变化特征,经矿场试验验证与其他方法相比有明显优势,为微量漏失监测和预测提供一种可行的方法,对油气钻井井漏风险的防控具有重要指导意义。
When lost circulation happens in the process of well drilling,the existing intelligent lost circulation monitoring methods can hardly obtain the long-term characteristics of data sequence and thus cannot realize the timely monitoring and prewarning of minor lost circulations,causing more serious lost circulation.In order to address this problem,this paper proposes a DCC-LSTM based intelligent minor lost circulation monitoring method which takes the advantage of the characteristic mapping capacity of Dilated and Causal Convolution(DCC)network and the sequential characteristic extraction capacity of Long Short-Term Memory(LSTM)network.This method makes up for the shortage of LSTM in long-term memory attenuation,and realizes the accurate monitoring and prediction of minor lost circulation.And the following research results are obtained.First,in the DCC-LSTM based intelligent minor lost circulation monitoring model,the long-term characteristics of monitoring parameters are extracted by using the DCC network and then mapped into a short data sequence,and the long-term variation trend of monitoring parameters is obtained by applying the LSTM network to process the characteristic short data sequence,so that the accurate monitoring of minor lost circulation is realized.Second,the optimal number of layers in the network can be obtained by using the method for determining the number of layers in the DCC network.The new structure of DCC network can reduce the long-term sequential trend information forgotten by LSTM by 24%.Third,compared with other lost circulation monitoring methods,the DCC-LSTM network can monitor early minor lost circulations accurately,with the advanced prewarning time increasing by 26 minutes,the monitoring accuracy rate increasing from 96.9%to 99.4%,and the false alarm rate decreasing from 6.4%to 1.1%.In conclusion,this method can acquire the long-term trend variation characteristics of monitoring parameters,and the field test demonstrates its obvious advantages over other methods.It provides a feasible method for the monitoring and prediction of minor lost circulation,and is of great significance to guiding the prevention and control of lost circulation risk during well drilling.
作者
孙伟峰
卜赛赛
张德志
李威桦
刘凯
戴永寿
SUN Weifeng;BU Saisai;ZHANG Dezhi;LI Weihua;LIU Kai;DAI Yongshou(College of Oceanography and Space Informatics,China University of Petroleum-East China,Qingdao,Shandong 266000,China;College of Control Science and Engineering,China University of Petroleum-East China,Qingdao,Shandong 2660000,China)
出处
《天然气工业》
EI
CAS
CSCD
北大核心
2023年第9期141-148,共8页
Natural Gas Industry
基金
国家自然科学基金项目“基于深度学习的深地叠前时空域地震子波提取方法研究”(编号:42274159)。
关键词
井漏
微量漏失
长时趋势特征
智能监测
扩张因果卷积网络
长短期记忆网络
Lost circulation,Minor loss
Long-term trend characteristics
Intelligent monitoring
Dilated and Causal Convolution(DCC)network
Long Short-Term Memory(LSTM)network
