With the steady rising of oil price in recent years, the depth of welldrilling and drilling costs increases dramatically. How to control drilling costs for drilling enterprises is very important. Welldrilling activities are special and complex. Different segments, types of well and technologies affect drilling costs differently. It is very useful for drilling enterprises to estimate accurately by analyzing the effective factors on drilling costs.Support vector machine(SVM) is a new generation of statistical-learning-theory-based machine learning technology which can solve the regression forecasting with small sample effectively by the replacement of structural risk minimization principle to empirical risk minimization principle.This dissertation firstly describes the theoretical bases of SVM-statistical learning theory (SLT). It focuses on the VC dimension of learning theory, the boundary of generalization, and structural risk minimization. It also introduces the fundament idea and algorithm of SVR.Then it introduces the welldrilling costs and the welldrilling work flow combining with the drilling data. Meanwhile, it analyzes the main elements affecting drilling costs. It also describes the steps of SVR applied in drilling costs forecasting, and it uses Support vector regression to establish a cost forecast model with drilling data accomplished by a drilling engineering company, which the model's forecast precision is highest by contrasting multiple linear regression and BP neural network with support vector regression.Next it describes the parameters of the search method and directly determine method. A new parameter searching method is proposed for welldrilling costs data characteristics. Simulation experiments based on standard data set and drilling costs data set reveal that the proposed method is an effective approach for SVR parameters selection with good performance.Finally it designed the welldrilling costs forecast system base on support vector machine theory and drilling of the actual situation of the enterprise, the system has achieved good results in practice. |