基于一维卷积神经网络的钻井周期预测

Drilling cycle prediction based on one-dimensional convolutional neural network

吴玉林 姜莹 程光华 马佳 钱育蓉

新疆大学软件学院，新疆 乌鲁木齐 830046 中国海油集团能源经济研究院，北京 100013 中国海洋石油集团有限公司，北京 100010

College of Software, Xinjiang University, Urumqi 830000, China CNOOC Energy Economics Research Institute, Beijing 100013, China China National Offshore Oil Corporation, Beijing 100010, China

海洋钻井工程投资和风险巨大，准确预测钻井周期和评估钻井风险有助于油田公司合理规划投资预算。鉴于现有的钻井周期预测大多采用概率统计学方法，文中利用神经网络在非线性关系拟合方面的优越性，结合卷积神经网络局部感知的特性，提出通过一维卷积神经网络（1DCNN）预测钻井周期。针对钻井事故对钻井周期的影响，提出对钻井事故进行量化分析，以事故量化、开钻年份、完钻井深、各井段顶深和钻深、各井段钻头尺寸和套管尺寸作为模型输入，建立事故井钻井周期预测模型，平均绝对百分误差和可决系数分别为11.66%，95.22%。根据事故量化分析结果，筛选无事故井，建立在无事故量化输入情况下的无事故井钻井周期预测模型。研究表明，事故井钻井周期预测模型有利于评估钻井风险，无事故井钻井周期预测模型可在新井不考虑钻井事故影响下提供较为准确的周期预测参考，对海上钻井周期预测具有一定的理论和实践意义。

The investment and risks of offshore drilling projects are huge. Accurately predicting the drilling cycle and evaluating drilling risks can help oilfield companies plan their investment budgets reasonably. In view of the fact that most of the existing drilling cycle prediction methods use probability and statistics methods, using the advantages of neural network in nonlinear relationship fitting, combined with the local perception characteristics of convolutional neural network, a one-dimensional convolutional neural network (1DCNN) is proposed to predict the drilling cycle. Aiming at the influence of drilling accidents on the drilling cycle, a quantitative analysis of drilling accidents is proposed. Using accident quantification, drilling year, total depth, top depth and drilling depth of each well section, drill bit size and casing size of each well section as models input, establish the drilling cycle prediction model of accident wells, which mean absolute percentage error and the coefficient of determination are 11.66% and 95.22%, respectively. According to the accident quantification results, accident-free wells are screened and the drilling cycle prediction model of accident-free well is established. The research shows that the accident well drilling cycle prediction model is beneficial to assess the drilling risk, and the accident-free well drilling cycle prediction can provide more accurate reference for cycle prediction without considering the influence of new well drilling accidents, which has certain theoretical and practical significance for offshore drilling cycle prediction.

10.6056/dkyqt202303018