Application of multi-output least-squares support vector machines (MLS-SVM) for classification of oil reservoirs from well testing signals

Analysis of well testing signals is a widely-used technique for characterizing the hydrocarbon reservoirs. This technique can simply reveal wellbore, reservoir, and boundary types through identification of characteristic shapes on the pressure derivative (PD) graphs. Since the traditional matching processes often fail to correctly detect characteristic shape of the PD graphs, in this study multi-output least-squares support vector machines (MLS-SVM) is proposed for identification two different reservoirs and four boundary types. Indeed, homogenous and dual porosity reservoirs with different external boundaries including closed, constant pressure, infinite acting, and single sealing fault are considered. Parameters of the MLS-SVM is firstly adjusted by 784 synthetic PD graphs obtained from PanSystem software. Performance of the designed MLS-SVM is then evaluated using an actual field and 196 new synthetic well testing signals. Classification accuracy is used for evaluation performance of the proposed smart model. Results indicates that the proposed smart approach is able to identify different reservoir and boundary types with 100% classification accuracy.

کلیدواژه ها:

Well testing signal ، pressure derivative plots ، MLS-SVM ، identification of reservoir and boundary types