Machine Learning Prediction of the Lost Circulation Events at the Well Planning Stage

Abstract Well construction process is often accomplished by multiple challenges, where lost circulation is one of the having most sifnificant impact. The downhole losses, partial or total, in many cases associated with karst, fractures, caverns or large voids, which cannot be identified at the well design stage with standard practices. The manuscript provides the detailed approach how the machine learning has been used to predict the loss circulation events at the planning stage. The novel approach is based on the 3 main pillars: data preparation, the machine learning itself and the model application. The data preparation includes some manual work, when the attributes of the offset wells shall be analyzed – daily reports (over 80 million sentences, in this specific case), including the losses depths; the trajectories, and the identification of lost circulation events on them; the microresistivity and caliper logs, recorded both across lost circulation zones and competent formations; and the seismic attributes, total 15 were picked. As a result, the 3D points with the attributes and labels for lost circulation events were created, where the Machine Leaning (ML) was applied. ML part includes the split of all the wells in 66/33 proportion, with study been performed in 66% of the wells, with further tune of the results. As the result the machine learning is applied to the seismic attributes and the trajectories to determine the risks of the lost circulation events. The model shows a low specificity in predicting the high probability of the lost circulation events, but a high specificity in predicting locations with a low probability of the lost circulation. Using the proposed approach, it is possible now to plan a well trajectory in such way, that probability of the losses are reduced by up to 95%. The proposed workflow can be incorporated in the entire well construction process from the start of a location preparation, to ensure the new trajectory (without losses) can reach the target zone. As well it helps at the planning stage to allocate the resources and material required to cure the expected losses or to execute the blind drilling.