Separating Well Log Data to Train Support Vector Machines for Lithology Prediction in a Heterogeneous Carbonate Reservoir

The prediction of lithology is necessary in all areas of petroleum engineering. This means that todesign a project in any branch of petroleum engineering, the lithology must be well known. Supportvector machines (SVM’s) use an analytical approach to classification based on statistical learningtheory, the principles of structural risk minimization, and empirical risk minimization. In thisresearch, SVM classification method is used for lithology prediction from petrophysical well logsbased on petrographic studies of core lithology in a heterogeneous carbonate reservoir in southwesternIran. Data preparation including normalization and attribute selection was performed on the data. Wellby well data separation technique was used for data partitioning so that the instances of each wellwere predicted against training the SVM with the other wells. The effect of different kernel functionson the SVM performance was deliberated. The results showed that the SVM performance in thelithology prediction of wells by applying well by well data partitioning technique is good, and that intwo data separation cases, radial basis function (RBF) kernel gives a higher lithology misclassificationrate compared with polynomial and normalized polynomial kernels. Moreover, the lithologymisclassification rate associated with RBF kernel increases with an increasing training set size.

Keywords:

Lithology Prediction , Support Vector Machines , Kernel Functions , Heterogeneous Carbonate Reservoirs , Petrophysical Well Logs