基于深度卷积特征重构的井漏事故预测

Lost Circulation Accident Forewarning Based on Deep Convolutional Feature Reconstruction Network

面向海上石油钻井井漏事故预测需求,以中国海油乐东10-1油田监测数据构建样本,利用ReliefF算法提取事故关键参量,并计算滑窗内积获取时序特征矩阵,以突出表达事故时序变化及关联耦合特征。设计并搭建深度卷积特征重构网络,训练学习正常钻井的特征表达规律,在测试阶段基于网络模型的特征重构误差捕捉事故前的征兆异常,实现井漏事故预测。结合钻井实际工程需求,研究并定义事故预测准确率和虚警密度指标评价算法结果,实验验证表明所提方法对井漏事故的预测准确率为75%,且虚警率低,表现优于自动编码器等常规算法,可为海上石油事故预警应用提供技术支撑,指导钻井作业安全生产。

This paper focuses on the forewarning of engineering accident in offshore oil drilling,and take the lost circulation accident in Ledong oilfield of China as example.First of all,the key parameters were extracted based on the ReliefF algorithm from the actural drilling monitoring data.Secondly,the sliding window inner product was calculated to construct the sequential feature matrix in order to highlight the time series change and correlation coupling information.Then,a deep convolutional feature reconstruction network was designed and constructed to learn the expression of the sequential feature matrix in the accident-free stage.Next,the anomaly symptom before the labeled accident was captured to forewarning the lost circulation accident based on the feature reconstruction error of the network in the test phase.Finally,the algorithm was evaluated and compared with the auto encoder by the accuracy of accident and false alarm density.The experimental results show that the method proposed in this paper is effective and can be applied to the accident forewarning and guide the engineering of offshore oil drilling.