内蒙古民族大学学报（自然科学版）2025，Vol.40Issue(5)：42-50,9.DOI:10.14045/j.cnki.15-1220.2025.05.007
基于迁移学习的钻井溢流预测方法
Drilling Overflow Prediction Algorithm Based on Transfer Learning
摘要
Abstract
Overflow are common and complex conditions in oil drilling that can lead to well control risks.To address the challenge of rapidly and efficiently identifying overflow,a domain adaptation method based on transfer learning is proposed,which can detect blowouts 10 minutes in advance with a small sample size.Features of oil drilling data are processed using variance selection and mutual information methods,and sample labels are con-structed using a sliding time window.A CNN-LSTM basic model is established.Additionally,multi-kernel MMD in the DAN method is applied to adjust the last three fully connected layers of the CNN-LSTM model,adapting the data of source and target domain.To prevent overfitting,Gaussian white noise is added for data augmentation.The experimental results show that the model achieves a blowout prediction accuracy in target domain data reached 78.98%,achieving effective early prediction.关键词
石油钻井/溢流预测/时间序列预测/迁移学习Key words
oil drilling/overflow prediction/time series prediction/transfer learning分类
信息技术与安全科学引用本文复制引用
赵孟雪,时小虎,边婧,付加胜,韩霄松..基于迁移学习的钻井溢流预测方法[J].内蒙古民族大学学报（自然科学版）,2025,40(5):42-50,9.基金项目
吉林省科技发展计划项目(20220201145GX) （20220201145GX）
