Prediction of Filtration-capacItance Properties of Terrigenous Reservoirs Using Factor Analysis and Neural Networks at an Offshore Field off Sakhalin Island
E.I. Korytkin 1,2, G.M. Mitrofanov 1,3
1 Trofimuk Institute of Petroleum Geology and Geophysics, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia
2 SakhalinNIPI Nefti I Gaza LLC, Yuzhno-Sakhalinsk, Russia
3 Novosibirsk State University, Novosibirsk, Russia
Keywords: 3D seismic exploration, factor analysis, multivariate regression analysis, neural networks, reservoir properties forecasting.
Abstract
The problem of forecasting the filtration-capacitance properties (FCP) of target horizons (porosity, effective thickness, oil saturation) using methods capable of learning on large volumes of heterogeneous data is considered. 3D seismic data, both full-stack (seismic attributes extracted from a seismic cube) and partial-stacks (synchronous inversion), are used, limited to the target horizons interval, obtained at one of the offshore fields in the north of Sakhalin. The large data volume and the strong interdependence of the extracted features required dimensionality reduction of the seismic parameters. The novelty of the developed approach lies in the application of factor analysis prior to performing FCP forecasting. As a result, 92% of the data variations associated with reservoir properties were represented by five main factors. Subsequent application of neural networks to the forecasting problem demonstrated that the identified factors are significant predictors of porosity, oil saturation, and effective thickness. Cross-validation identified zones with low accuracy in FCP estimation – resulting from the strong influence of gas on seismic data – as well as zones with high accuracy in FCP estimation, which were subsequently used to determine optimal weights and the final forecasts. Based on the results of FCP forecasting using factor analysis and neural networks, correlation coefficients at test well locations ranged from 0.92 to 0.93, and forecast accuracy – verified following the drilling of two production wells – fell within the 85–93% range, thereby demonstrating the effectiveness of the proposed technology.
|
