Prediction of Reservoir Parameters using Support Vector Machines, A Machine Learning Approach

Reservoir parameters such as effective porosity, clay volume, and water saturation are essential in estimating the hydrocarbon reserves in place. Well logs are continuous measurements of the physical properties of the formation. Quantitative interpretation of well logs coupled with core calibration can give reservoir parameter estimates, which can be good approximates to that of core results. Thus, well log analysis is a practical alternative to the expensive and time-consuming core studies. Various core-based non-linear mathematical models are used for this purpose.
Support vector machines regression (SVM) is a supervised machine learning algorithm, which can be used for regression and classification problems. A support-vector machine uses kernel trick to constructs a set of hyperplanes in a high- or infinite-dimensional feature spaces of the inputs to solve the non-linear regression problems. In the present study, the application of SVMs in estimating the clay volume and effective porosity from well logs are examined. First, the algorithm is applied to synthetic data to test the applicability and parameter optimization. Further SVM is trained and tested on siliciclastic formations and found acceptable results. The root mean squared error between SVM predicted and manually interpreted results is for VCL is 0.007 and PHIE is 0.005.