A clustering-guided feature enhancement and Bayesian-optimized LSTM approach for LC temporal anomaly detectionOA
Lost circulation(LC),as one of the high-risk accidents in drilling,usually occurs when the wellbore pressure is greater than the formation pressure,causing a large amount of drilling fluid to seep into the formation and resulting in a significant decrease in the flow rate at the wellhead.Currently,existing LC prediction studies mainly rely on logging parameters for time series prediction,ignoring the problem of the inherent imbalance in the ratio of positive and negative samples in the LC dataset and geological information.To address these issues,this study proposes a clustering-guided feature enhancement and Bayesian optimization Long Short-Term Memory(BO-LSTM)method for LC time series anomaly detection.This method,based on the mud logging and geological structural information of faults and lithology risks,leverages clustering algorithms to uncover the latent structural information within samples to construct enhanced features,which combine with the selected key features,forming a joint representation that is fed into the LSTM model.At the same time,the BO algorithm is introduced to adaptively optimize the LSTM hyperparameters,and finally outputs the LC occurrence''s probability.The results show that the proposed method performs well in LC prediction,with an average false negative rate of 0.825%,an average false positive rate of 1.525%,and an average lead time of 4.825 min for effective early warnings,significantly improving the accuracy and timeliness of early LC identification.
Yating Lei;He Zhang;Haibo Liang;Yihui Han;Jialing Zou
Southwest Petroleum University,Chengdu 610500,ChinaSouthwest Petroleum University,Chengdu 610500,ChinaSouthwest Petroleum University,Chengdu 610500,ChinaThe 10th Research Institute of China Electronics Technology Group Corporation,Chengdu,610036,China Agile and Intelligent Computing Key Laboratory of Sichuan Province,Chengdu,610036,ChinaSouthwest Petroleum University,Chengdu 610500,China
能源科技
LC early warningLong short-term memory networkBayesian optimizationClusteringFeature enhancement
《Petroleum》 2026 (3)
P.509-522,14
supported by the National Natural Science Foundation of China(Grant No.52404005)the Innovation Research Community Project of the Natural Science Foundation of Sichuan Province(Grant No.2024NSFTD0053).
评论