TY - JOUR

T1 - Is Support Vector Regression method suitable for predicting rate of penetration?

AU - Kor, Korhan

AU - Altun, Gursat

N1 - Publisher Copyright:
© 2020 Elsevier B.V.

PY - 2020/11

Y1 - 2020/11

N2 - Drilling operations constitute the major part of the exploration costs. Drill bits are the primary part required to be changed frequently due to its quick wearing nature. In order to reduce the drilling cost, rate of bit penetration (ROP) must be estimated or optimized by using several drilling parameters. One of the most commonly used methods for estimating the optimum penetration rate is Bourgoyne and Young Method (BYM) that is eight-parameter model. However, mathematically speaking, minimum of thirty different sets of data zones, which must be taken from uniform-lithology sections (particularly shales) are needed to get accurate results. To construct the functional relationship with the data, a multiple regression analysis is performed. When there is not enough data, the accuracy of BYM decreases. Support Vector Regression (SVR), one of machine learning methods, is commonly used to increase the prediction accuracy when insufficient data is available. In this study, two kinds of regression techniques are used to predict ROP: multiple linear regression (BYM) and SVR. For the calculations, two different field datasets obtained from the literature are used. Each dataset is trained and tested individually and completely with respect to all data points, using different scenarios. The results of different predicting methods for each dataset are compared in a statistical point of view to demonstrate the effectiveness of the SVR method subjected to insufficient input data. The results show the significant effect of data selection on the accuracy of ROP prediction. SVR is used for ROP prediction for the first time by fully implementing in BYM. This study paves the way for predicting ROP when the data is insufficient.

AB - Drilling operations constitute the major part of the exploration costs. Drill bits are the primary part required to be changed frequently due to its quick wearing nature. In order to reduce the drilling cost, rate of bit penetration (ROP) must be estimated or optimized by using several drilling parameters. One of the most commonly used methods for estimating the optimum penetration rate is Bourgoyne and Young Method (BYM) that is eight-parameter model. However, mathematically speaking, minimum of thirty different sets of data zones, which must be taken from uniform-lithology sections (particularly shales) are needed to get accurate results. To construct the functional relationship with the data, a multiple regression analysis is performed. When there is not enough data, the accuracy of BYM decreases. Support Vector Regression (SVR), one of machine learning methods, is commonly used to increase the prediction accuracy when insufficient data is available. In this study, two kinds of regression techniques are used to predict ROP: multiple linear regression (BYM) and SVR. For the calculations, two different field datasets obtained from the literature are used. Each dataset is trained and tested individually and completely with respect to all data points, using different scenarios. The results of different predicting methods for each dataset are compared in a statistical point of view to demonstrate the effectiveness of the SVR method subjected to insufficient input data. The results show the significant effect of data selection on the accuracy of ROP prediction. SVR is used for ROP prediction for the first time by fully implementing in BYM. This study paves the way for predicting ROP when the data is insufficient.

KW - Drilling optimization

KW - Machine learning

KW - Multiple linear regression

KW - Rate of penetration prediction

KW - Support vector regression

UR - http://www.scopus.com/inward/record.url?scp=85087126972&partnerID=8YFLogxK

U2 - 10.1016/j.petrol.2020.107542

DO - 10.1016/j.petrol.2020.107542

M3 - Article

AN - SCOPUS:85087126972

SN - 0920-4105

VL - 194

JO - Journal of Petroleum Science and Engineering

JF - Journal of Petroleum Science and Engineering

M1 - 107542

ER -