Rock Typing Mapping Methodology Based on Indexed and Probabilistic Self-Organized Map in Shushufindi Field

Abstract

Abstract The Shushufindi field is located in the Oriente basin of Ecuador. The field was discovered in 1972 and widely developed with about 247 wells covering an area of approximately 400 km2. The implementation of lithofacies characterization in 98% of the existing wells has given a reliable description in about 92% of the wells in the current geomodel, which demonstrates, the validity of the deterministic method. A robust petrophysical rock type (PRT) classification can significantly improve the chances of success for all wells, focusing on layered reservoir rocks recognized as the major energy resource in recent years. The vertical and lateral classification of rock heterogeneity in the form of rock types is critical to understand the flow dynamics of the reservoirs. Well logs are the best option for formation evaluation as they provide high vertical resolution measurements. However, rock type's classification using only well logs interpretation techniques, has its limits. In this paper, we introduce a rock type neural network technique based on Indexed and Probabilistic Self-Organized Mapping (IPSOM) which was designed for the geological interpretation of well log data, facies prediction and optimal derivation of petrophysical parameters. The rock typing was based on cored wells in a 3-step approach. Preliminary rock type identification was based on sedimentology description and routine core analysis. In parallel, it was refined with high pressure mercury injection data to describe accurately the porous media. The porosity and permeability ranges were established to elaborate a sand facies classification represented by Petrophysical Rock Type through Winland method. The neural network was first trained on cored reservoirs, and then propagated to uncored wells using the classification model relationship with electrical logs. Finally using the IPSOM classification model, a permeability-porosity relationship for each rock type was obtained, providing input to the dynamic model to predict and validate permeability. This paper present a reservoir characterization enhancement technique using neural network, which has proven its utility in refining the dynamic model of the Shushufindi field and directly contributing to the operator by improving production from layered reservoirs.