摘要
目前的人工神经网络和贝叶斯网络等方法无法连续长时间准确早期预测溢流。为此,针对钻井现场数据特点,提出基于CNN-LSTM融合网络的溢流工况预测深度学习方法,结合溢流的关键参数变化,融合多特征数据尽早发现溢流征兆,从而实现溢流工况早期预测。将采集的所有钻井特征参数用于模型训练和溢流预测,准确率高于单独的CNN结构或单独LSTM结构。为进一步提高模型效率和使用较少现场参数,有针对性地筛选了与溢流紧密相关的一些钻井参数进行试验,结果表明采用CNN-LSTM融合的网络结构优于单独的CNN或单独LSTM结构,预测准确率可达到89.55%。对某钻井数据的预测分析结果表明,该方法能够提前10 min准确预测溢流的发生,可为后续采取相应措施争取到宝贵时间。
The current methods such as artificial neural network and Bayesian network cannot continuously predict the overflow accurately in early stage for a long time.In this paper,based on the characteristics of rig site data,a CNN-LSTM fusion network based deep learning method for predicting overflow behavior was proposed;combined with the change of key parameters of overflow,fusing multi-feature data,the overflow symptoms were discovered as early as possible,thus realizing the early prediction of overflow behavior.All the collected drilling characteristic parameters were used for model training and overflow prediction,showing a higher accuracy than that of single CNN structure or single LSTM structure.In order to further improve the model efficiency and use fewer rig site parameters,some drilling parameters closely related to the overflow were selected to perform test,showing that the CNN-LSTM fusion network structure is better than single CNN structure or single LSTM structure,and the prediction accuracy reaches 89.55%.The prediction analysis results of a certain drilling data show that this method can accurately predict the occurrence of complex overflow by 10 min in advance,which wins valuable time for taking corresponding measures subsequently.
作者
付加胜
刘伟
韩霄松
李丰欣
Fu Jiasheng;Liu Wei;Han Xiaosong;Li Fengxin(CNPC Engineering Technology R&D Company Limited;Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education;College of Computer Science and Technology,Jilin University)
出处
《石油机械》
北大核心
2021年第6期16-22,共7页
China Petroleum Machinery
基金
国家自然科学基金重大项目“天然气水合物钻采井筒多相流动障碍形成机制与安全控制方法”(51991363)
中国石油天然气集团有限公司直属院所基础研究和战略储备技术研究基金项目“智能化钻井设计与井下复杂监控模型与方法研究”(2019D-5008-03)。
