A machine-learning based quantitative evaluation of the fluid components on T2-D spectrum

Abstract Estimating the saturation of different fluids in reservoirs is key to determine the reservoir quality and to calculate reserves. Due to the overlap of different fluids on the inverted T2-D spectrum, estimating the saturation and evaluating the fluid components proves to be a difficult task. In this paper, we propose a machine learning method that integrates GABP neural network and Gaussian Mixture Model (GMM) to quantitatively evaluate the fluid components on T2-D spectrum. Distinct from other machine learning approaches in oil and gas exploration, the proposed method accounts for the physical significance between the T2-D spectrum and fluid saturation to determine a new saturation formula. We perform numerical simulations and subsequently process the simulated T2-D data. The results demonstrate the greater accuracy and stability of the GABP predictions compared to the BP neural network. However, both sets of results have no physical significance. The GMM is able to determine different Gaussian probability distribution functions (PDFs), which are then used to characterize different fluids on T2-D spectrum. The saturation determined via the Gaussian PDFs is more accurate than those of the GABP and BP. Our work reveals the ability of the proposed method to quantitatively evaluate the fluid components on T2-D spectrum.