一种基于2D-CNN深度学习的钻井事故等级预测新方法  

A new method for predicting drilling accident level based on 2D-CNN deep learning

作　　者：赵春兰[1] 屈瑶 王兵[2] 范翔宇[3] 赵鹏斐 李屹 何婷 ZHAO Chunlan;QU Yao;WANG Bing;FAN Xiangyu;ZHAO Pengfei;LI Yi;HE Ting(School of Science,Southwest Petroleum University,Chengdu,Sichuan 610500,China;School of Computer Science,Southwest Petroleum University,Chengdu,Sichuan 610500,China;State Key Laboratory of Oil&Gas Reservoir Geology and Exploitation//SouthwestPetroleum University,Chengdu,Sichuan 610500,China;School of Earth Science and Technology,Southwest Petroleum University,Chengdu,Sichuan 610500,China)

机构地区：[1]西南石油大学理学院 [2]西南石油大学计算机科学学院 [3]“油气藏地质及开发工程”国家重点实验室·西南石油大学 [4]西南石油大学地球科学与技术学院

出　　处：《天然气工业》2022年第12期95-105,共11页Natural Gas Industry

基　　金：国家自然科学基金面上项目“电磁振荡条件下成岩天然气水合物热动力学特性及对井壁稳定的影响研究”(编号:51774246);“深部火山岩体井周多尺度力学失稳机理及钻井低力学扰动点的地层空间分布规律研究”(编号:42172313);四川省自然科学基金项目“基于大数据技术的致密砂岩气藏剩余气定量分级评价系统”(编号:2022NSFSC0283)。

摘　　要：鉴于钻井安全事故分级风险评价过程中,存在安全事故风险指标较少且多为2分类预测的实际问题。为此,在利用模糊C均值算法确定钻井事故等级的分类的基础上,根据信息增益值对多维事故风险指标进行一次降维;进而将降维后的风险指标作为模型输入,由卷积层提取事故特征,池化层进行二次降维,构建双层2D-CNN的事故等级预测模型,最后通过激活函数(Softmax)判断钻井事故等级,提出一种基于二维卷积神经网络(2D-CNN)的钻井事故等级预测的新方法。研究结果表明:①较之于其他方法,新方法经过两次降维将多维钻井事故指标由73维降低至4维,降低模型计算复杂度;②不同于钻井事故发生与否的二分类问题,根据事故的严重程度划分成四种事故等级,以实现多分类预测;③现场应用效果表明,新方法的准确率为91.7%,损失值为0.409,预测效果优于BP神经网络模型和1D-CNN模型。结论认为,新方法能较好地将现场作业数据用于钻井事故等级的预测,对于钻井事故风险分级评价具有广泛应用和推广价值。For risk assessment of drilling accidents, there are only few risk indicators, which are mostly based on binary classification.This paper presents a new method for predicting drilling accident level based on two-dimensional convolutional neural network(2D-CNN).First, the fuzzy C-means(FCM) algorithm is used to determine the classification of drilling accident levels, and the first dimensionality reduction is performed on multi-dimensional risk indicators according to the information gain value. Then, a two-layer 2D-CNN accident level prediction model is constructed, with the risk indicators after dimensionality reduction as the input of the model, the convolutional layer for extracting the accident feature, and the pooling layer for second dimensionality reduction. Finally, the drilling accident level is judged by the activation function(Softmax). The research results show that, compared with other methods, the new method reduces the multi-dimensional risk indicators from 73 to 4 dimensions after two dimensionality reductions, which reduces the computational complexity of the model. Different from the binary classification based on whether a drilling accident occurs or not, the new method divides the drilling accidents into four levels according to the severity to achieve multiple classifier prediction. Field application demonstrates that the new method provides an accuracy of 91.7% and a loss value of 0.409, indicating a better result than the BP neural network and the 1D-CNN models. In conclusion, the new method can effectively use the field operation data to predict the drilling accident level, and it is worthy of promotion to the classification and risk assessment of drilling accidents.

关 键 词：多维钻井事故 事故等级 多分类预测 深度学习 二维卷积神经网络 模糊C均值算法 信息增益 

分 类 号：TE28[石油与天然气工程—油气井工程] TP18[自动化与计算机技术—控制理论与控制工程]