Forecasting the drilling parameters is very relevant. Trained machine learning models make it possible to obtain a predictive value of regime parameters using previously accumulated experience without complex calculations. The applying of modern tools requires new approaches to the analysis of accumulated data. The main problem is the marking of possible cases for forecasting. The data were collected in the form of values of drilling parameters generalized over the intervals by the measured depth. A machine learning model was built to determine equivalent circulation density (ECD) based only on the project values of the well design and geology. This forecast makes it possible to determine the segments where the actual ECD deviated from the expected one. Thus, the data were re-labeled and the targets of the forecast were determined. Machine learning can classify these segments. Interpretations are discussed.
Any special tasks for machine learning impose restrictions on the data used. The article discusses the features of drilling data and the limitations of choosing such data for training. The problem of forecasting special cases is considered. Segments were identified where it was necessary to increase the ECD due to high gas readings by increasing the density of the solution. The drilling parameters of previous drilling intervals were taken as predictors. Algorithms trained specifically on such cases are able to predict them in the future. A positive forecast means a possible need to increase the mud density in the next drilling interval. The algorithm distinguishes such states from those when the density of the fluid remains unchanged. Difficulties in interpreting and evaluating such forecasts associated with working on actual data are considered.
1. Vadetskiy Yu.V., Spravochnik buril’shchika (Driller’s handbook), Moscow: Akademiya Publ., 2008, pp. 209–249.
2. Alkinani H.H., Al-Hameedi A.T.T., Dunn-NormanS., Lian D., Application of artificial neural networks in the drilling processes: Can equivalent circulation density be estimated prior to drilling, Egyptian Journal of Petroleum, 2020, V.29, pp. 121–126, DOI: https://doi.org/10.1016/j.ejpe.2019.12.003
3. Rooki R., Application of general regression neural network (GRNN) for indirect measuring pressure loss of Herschel–Bulkley drilling fluids in oil drilling, Measurement, 2016, V.85, pp. 184–191, DOI: http://doi.org/10.1016/j.measurement.2016.02.0374. Yu Y., Liu Q., Chambon S., Hamzah M., Using deep kalman filter to predict drilling time series, Proceedings of International Petroleum Technology Conference, March 2019, DOI: https://doi.org/10.2523/IPTC-19207-MS