Destek vektör regresyonu yöntemi ile ilerleme hızı optimizasyonu

Drilling operations constitute the major part of the exploration costs. During operations, drill bits are the primary part needs to be changed frequently due to its quick wearing nature. In order to reduce the drilling cost, the optimum bit pulling time must be determined. To determine the optimum bit pulling time, either rate of penetration or the tooth wearing parameter must be estimated. The most common method that developed for estimating the optimum time for bit change is "Bourgoyne and Young" method. In this method, eight parameter coefficients are needed. To obtain these coefficients, thirty different data which can be taken from either different shale zones inside thirty different wells in a field or thirty different shale points from one well is needed. However, when there is not enough data taken from thirty different shale sections, the accuracy of "Bourgoyne and Young" method decreases. To construct the functional relationship with the data and parameter coefficients, a regression analysis must be performed. In this study, two kind of regression technique is used and the results are compared to each other. First technique is the multiple regression analysis, which is also used in "Bourgoyne and Young" method. This analysis applies least-squares-principled-regression to the data and calculates the parameter coefficients in order to estimate the target function. The second technique is one of different types of machine learning algorithms, called Support Vector Regression. In this technique, first, the data is divided into train and test datasets. Then, the regression model is constructed by using train datasets. At last, the model is applied to test datasets in order to predict the target values for the function. For the calculations, the selection of training and testing data sets are divided into cases with different scenarios. The results of different predictor methods for each scenario are compared with each other in the corresponding case. The results show the significant effect of data selection on the accuracy of penetration rate prediction. One of the most powerful methods in machine learning, Support Vector Regression, is used for rate of penetration optimization for the first time in the literature with this thesis study. In this way, the chance for further investigations and studies on the practicability of Support Vector Regression on penetration rate optimization is created.