Logging Data And Support Vector Machine-based Rock Drillability Study

Drillability has important significance for drilling engineering and has a direct impact on the determination of drill bit type selection and drilling rate forecasts, the major oil fields at home and abroad for the determination of lithology are very seriously. However, suitable for the conditions of oil and gas wells drillability question had not yet completely resolved. On determination of rock drillability, most research methods used adjacent Wells data analysis to get through the laboratory. This method has great limitations and cost is higher to collect a lot of typical core more difficult, especially in Wells and spacing far fewer new exploration area. Therefore, in view of the new exploration area of the exploration area with core data, it needs to some new ways and new methods to get drillability.With the development of the theory of statistics, Most has already started to use machine learning theory to study the development of oil and gas exploration.However, traditional historical fitting methods has high cost and the poor effect. Compared with neural network, support vector machine can be effectively solved under the condition of limited samples high-dimensional data model building problems, it have good generalization ability,converge to the global optimal and dimension not sensitive, etc. At present, the support vector machine (SVM) method after the neural network has become the most popular machine learning field of the research direction. Therefore, this paper use support vector machine theory to research strata rock drillability.This paper firstly summarized statistical learning theory and intelligent algorithms introduced the principle of support vector machine regression,summarized several nuclear function selection methods,and then, introduced the rock drillability level the definition and some experimental methods to determine the value drillability level value, enumerated several different logging methods to calculate the formation parameters.Secondly, on the basis of the logging data, using logging parameter selection which influenced by model, it established the strata rock drillability model, and using the model of a well of the SC oilfield predicting actual test data. On support vector machine (SVM) model, it used several common respectively the kernel function to predict, and compared the fitting effect, so as to determine the optimal kernel functions. The selection of parameters in kernel function, used the grid search method to determine a set of parameters, according to the historical fitting adjusted steps to fit again until fitting the optimal effect. Finally, compared this method and artificial neural network, it showed this method is high and feasible.