Machine learning-based diagnosis of fracture-induced lost circulation and intelligent selection of lost circulation control fluids

Objective and Method  Drilling operations in the Changqing area face several challenges in lost circulation, including complex types, significantly different characteristics, and difficulty in designing plugging formulations. To address these issues, this study proposes a DSX-Hybrid coupled intelligent decision-making framework, which serves as an integrated scheme to identify lost circulation channels, perform intelligent selection of plugging formulations, and offer recommendations on construction process. In this framework, given the scarcity of labeled log data, the density-based spatial clustering of applications with noise (DBSCAN) is employed to mine latent geological characteristics of unlabeled data. Then, the self-attention convolutional neural network (SACNN) is constructed to integrate local features with long-range layer-wise correlations, thus enabling the semi-supervised prediction of fracture parameters. Lastly, the physics-constrained XGBoost model, combined with eight-dimensional characteristic parameters, is used to achieve the intelligent selection and performance prediction of plugging formulations. Concurrently, standardized construction procedures for lost circulation control are clarified.

Results and Conclusions  The diagnostic accuracy of lost circulation channels is increased from 73% to 94.6%. The semi-supervised model can make full use of massive unlabeled logging data and achieves more stable identification of fractures with various scales compared with traditional algorithms. The DSX-Hybrid model realizes accurate matching between fractures and particle sizes of plugging materials with the coefficient of determination (R²) over 0.945. The indoor plugging pressure bearing capacity of recommended formulas reaches 10.7 MPa, and the stable on-site pressure bearing capacity is 7.6 MPa, with a one-time plugging success rate of 100% in multiple blocks of Changqing Oilfield. This model provides a practical intelligent technical approach for lost circulation control in Changqing area. Further research can integrate real-time logging-while-drilling data, enrich datasets of extreme lost circulation conditions, and extend the application to other fractured oil and gas blocks.