Analysis of machine learning approaches for the interpretation of acoustic fields obtained by well noise data modelling

Assessing the phase composition of the fluid in a well based analysis of the frequencies of the radial resonance modes excited by acoustic noise in the inflow zone is a promising method for interpreting the results of passive noise metering. Machine learning makes it possible to take into account many factors affecting the spectrum of the measured signal, extracting from them exactly those factors associated with a change in phase composition. In order to build the best model, machine learning approaches such as linear regression with different variants of regularisation, Bayesian regression, neural net, methods of supporting vectors, decision tree, random forest and gradient boosting are considered. Data sets for training and testing the algorithm were obtained on the basis of scenarios calculated using a two-dimensional mathematical model with the different values of the bed parameters and ratio of volume fractions of the well filling fluids. The effect on the assessment accuracy of the phase composition of various factors, including the presence of acoustic device housing, the foreign noise in the signal and the shape of the signal spectrum, was checked. It is shown that in the absence of data distortion, it is possible to build models that provide an absolute error in the assessment of the phase composition about 1% after the zone of fluid inflow and about 5% in the zone before the inflow.

acoustic noise, interpretation, machine learning, linear regression, reference vector method, random forest, gradient boosting, neural net

1. Ipatov A.I, Kremenetskiy M.I. Geofizicheskiy i gidrodinamicheskiy kontrol’ razrabotki mestorozhdeniy uglevodorodov. Moscow, NITs «Regulyarnaya i khaoticheskaya dinamika», 2006, 780 p. (In Russian)

2. Nikolaev S.A., Ovchinnikov M.N. Generatsiia zvuka filtratsionnym potokom v poristykh sredakh. Akusticheskii zhurnal, 1992, Vol. 38, no 1, pp. 114–118. (In Russian)

3. Krasnovidov E.Yu. Sozdanie metodiki akustiko-gidrodinamicheskikh issledovaniy poristykh sred i skvazhin. Diss. kand. tech. nauk. Moscow, 2005, 171 p. (In Russian)

4. DiCarlo D.A., Cidoncha J.I.G., Hickey C. Acoustic measurements of pore-scale displacements. Geophysical Research Letters, 2003, Vol. 30, no 17, P. HLS 2-1–2-5. DOI: 10.1029/ 2003GL017811.

5. Mikhailov D.N., Sergeev S.I. Investigation parameters for sound induced by fluid displacement in rock samples. Water Resources Research, 2019, Vol. 55, no 5, P. 4220–4232. DOI: 10.1029/2018WR024168.

6. Marfin E.A., Gaifutdinov R.R., Metelev I.S. Investigation of the Intensity of Acoustic Emission During Gas Filtration Through Porous Media. Geomodel, Gelendzhik, 2018, DOI: 10.3997/2214-4609.201802405. (In Russian)

7. Mutovkin N.V., Mikhailov D.N., Sofronov I.L. Modeling of acoustic fields excided by fluid flow in near-wellbore zone. Mathematical Models and Computer Simulations, 2019, Vol. 31, no 6. pp. 95–106. DOI: 10.1134/S0234087919060066.

8. Mutovkin N.V., Mikhailov D.N., Sofronov I.L. Analysis of Modeling of Acoustic Fields Excited by the Flow Noise in the Formation Inflow Zones. Geomodel, Gelendzhik, 2018. DOI: 10.3997/2214-4609.201802388. (In Russian).

9. Mutovkin N.V., Mikhailov D.N., Sofronov I.L. Estimation of fluid phase composition variation along the wellbore by analyzing passive acoustic logging data. SPE Russian Petroleum Technology Conference, Moscow, 2019. ID: SPE-196845-MS. 15 p.

10. Hastie T., Tibshirani R., Friedman J.H. The elements of statistical learning: data mining, inference, and prediction. Springer Series in Statistics, 2009, 745 p.

11. Bishop M. Pattern recognition and machine learning. SpringerVerlag New York, 2006, 738 p.

12. Plyushchenkov B.D. Turchaninov V.I. 2000. Acoustic Logging Modeling by Refined Biot’s Equations. Int. J. Mod. Phys. C., 2000, Vol. 11, no 2, pp. 365–397.

13. Pedregosa F., Varoquaux G., Gramfort A. et al. Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research, 2011, Vol. 12, pp. 2825–2830.