Analysis of machine learning algorithms applicability for tasks of interpolation and geological properties forecasting within interwell space

Abstract

Obviously, in last years complexity of oil and gas fields has increased dramatically and the era of simple and easy retrievable resources has almost gone. This situation leads to necessity of careful and comprehensive uncertainty quantification and risk analysis during field evaluation and planning development strategy. Conventional and widely used geological modelling algorithms based on kriging and stochastic simulation process produce biased results due to not accurate variogram ranges induced by the lack of knowledge about geological setting of target formation. It is necessary to develop a tool that could explore available geological data and retrieve spatial environmental dependencies of a particular field and then exploit them for the following geological modelling. Modern statistical methods such as machine learning algorithms can be used for these tasks.
In this research applicability of machine learning algorithms for the task of interpolation and reservoir properties prediction within interwell space was analyzed. Robustness and quality of forecast produced by different machine learning models also were considered. Influence of information amount and sparsity of geological data on prognosis accuracy were estimated in order to determine range of used method applicability.