LEE- Petrol ve Doğal Gaz Mühendisliği Lisansüstü Programı
Bu topluluk için Kalıcı Uri
Gözat
Konu "Gas well drilling" ile LEE- Petrol ve Doğal Gaz Mühendisliği Lisansüstü Programı'a göz atma
Sayfa başına sonuç
Sıralama Seçenekleri
-
ÖgePenetration rate optimization in heterogeneous formations with support vector machines method(Lisansüstü Eğitim Enstitüsü, 2021) Kor, Korhan ; Altun, Gürşat ; 709835 ; Petrol ve Doğal Gaz MühendisliğiThe exploration of petroleum and natural gas resources is at topmost importance because of the unending demand for energy resources. Due to the oil companies' cost-efficient policy, the importance of reducing cost and increasing performance has ascended. Accordingly, there have been significant advances in drilling technology. Nowadays, cost and performance can be optimized thoroughly by using developing technology and computer science. The primary optimization problem in oil and gas exploration has always been related to minimize costs due to market volatility related to the decreasing oil prices. Drilling operations constitute a significant part of the exploration costs. Thus, the main objective of cost optimization is to reduce total well costs. One way to achieve this goal is to optimize or maximize the rate of penetration (ROP). The ROP starts to decrease as a drill bit wears within a run that cause additional cost. Practically, if the lithology is homogeneous, the optimum ROP can be achieved by adjusting controllable drilling parameters and considering the drilling cost starts to increase after a minimum value. On the other hand, this approach usually fails when drilling through complex lithological formations known as a heterogeneous environment distinctly non-uniform in lithological composition. There is no widely accepted model for defining the optimum ROP since various variables affect the cost and time relationship. It is clear that maximizing ROP or minimizing drilling costs is a multi-parameter optimization problem that requires optimization techniques. There exist some mathematical and statistical models to optimize ROP and various drilling parameters. The most commonly used ROP model in the drilling industry is Bourgoyne and Young Method (BYM). In BYM, the aim is to calculate eight regression coefficients related to standard drilling parameters by applying multiple linear regression (MLR) analysis to drilling data for predicting ROP. The BYM involves eight drilling parameters and requires statistically at least thirty inputs obtained from the formations with uniform lithology. The number of input data required to obtain such a large number of parameters may not always be possible. This is especially the case in drilling environments where there are not many shale zones, and where there are few wells in a field, or where complex lithology is dominant such as in the fields of Turkey. Thus, the results obtained by using the BYM in these limited conditions are meaningless, observations obtained from the field practice, and method results give very different values. On the other hand, the support vector machines (SVM) method effectively predicts ROP with higher sensitivity and conservation without decreasing the number of parameters. The main purpose of this study is to implement a different type of regression model as an alternative to multiple linear regression by modifying BYM by introducing several geomechanical parameters to estimate a relationship between the rock mechanical properties and the heterogeneity, which are shale index, uniaxial compressive strength, brittleness index, shear failure gradient, and torque. In this study, several predictive methods are used to build a model for ROP prediction: MLR, support vector regression (SVR), and artificial neural networks (ANN). Feature selection is performed via random forests method. The prediction accuracy of each predictive model is compared by using several statistical comparison criteria: root mean square error (RMSE), correlation coefficient, and statistical significance (p-value). The main focus of the applications in this study is to perform rate of penetration (ROP) predictions in heterogeneous formations. With a field data taken recently, it is possible to show the prediction performances of MLR, SVR, and ANN on the dataset taken from heterogeneous lithology. Since data quality is one of the essential process of data analysis. Hence, data cleansing and data selection are made precisely while constructing the cases and the datasets. In the results, it is clearly shown that BYM performs well in terms of ROP prediction in homogeneous formations. In heterogeneous formations, SVR with radial basis function (RBF) kernels gives better prediction results in terms of RMSE. On the contrary, for the homogeneous formations, linear models produce lower errors. ANN also performs well in heterogeneous formations in terms of ROP prediction. Adding some extra features related to the rock mechanics have different effects on prediction performances for all models. Torque is found as the key parameter to define the relationship between ROP and heterogeneity. Depth is another important parameter on ROP prediction in heterogeneous formations. Moreover, feature selection based on the random forests algorithm is applied. The cut-off value is found as formation-specific. Hence, it is stated that feature selection should be performed for each data set exclusively. The studies' findings in this dissertation are expected to provide significant time and cost savings since it is expected that faster and more accurate results will be achieved than those of currently available methods. The results will provide a basis for monitoring the condition of a drill bit on a real-time basis during any drilling operations and determining the best time to change the drill bit while avoiding a possible increase in the overall drilling cost.