Self-Organizing Maps for Lithofacies Identification and Permeability Prediction

Abstract

TX 75083-3836, U.S.A., fax 01-972-952-9435. AbstractMethods of Artificial Intelligence like Back-Propagation Neural Networks (BPNN) have become popular software tools to predict permeability and porosity from well logs during the last several years.Similar to Multiple-Linear Regression models, Back-Propagation Neural Networks are trained with a set of target values from core measurements.The Self-Organizing Map (SOM) Neural Network method applies an unsupervised training algorithm. Until now this approach has mainly been applied for clustering purposes only, not for predicting reservoir properties.In a new application, SOM technology has been merged with statistical prediction methods to derive the following types of information from well logs and core measurements in one step: 1) Synthetic lithofacies system (clustering) 2) Porosity and permeability (prediction) SOM technology also provides a data visualization tool which allows evaluating relationships between input variables (well logs) and output variables (reservoir properties).SOM models can also be combined with BPNN in order to subdivide the entire set of well log patterns into different lithofacies and run individual BPNN models for each facies.Application of this method showed increase in prediction accuracy and significant timesavings. This paper should be viewed and printed in color.

Bibliometric Analysis on Numerical Lithofacies Identification for Reservoir Characterization in the Period of 1980 -2021

“…The geological model of a hydrocarbon reservoir is based on reservoir properties estimation, such as lithology, porosity, permeability and fluid type [27] [28][29] [30]. The use of robust mathematical methods aims to reduce uncertainty when generating such a model, the more reliable results, so the more helpful to minimize risks and costs.…”

“…The artificial neural network has ability to model non-linear relationships between variables, which is highly adaptable to input datasets and does not require prior to statistical data distribution knowledge. The fundamental features place artificial neural networks among the most important clustering and classification methods in the last decade [32] [33], therefore, are mostly used in reservoir geology, which involves geological heterogeneity at various scales and assemblage treatments of data from several sources [34] [35][36] [37]. As an example, there are three basic uses of artificial neural networks to process data from well logs: i.e.…”

“…As an example, there are three basic uses of artificial neural networks to process data from well logs: i.e. permeability and/or porosity prediction, synthetic seismogram generation and electrofacies determination [38].…”

Remote Field Management of Greenfield Assets in Australia and East Timor: Delivering Value by Automated Production Surveillance and Optimization Platform

“…The recorded data at both fields was a subset of a large quantity of metered values, comprising pressure, temperature, and rate information at different locations throughout the well, gathering network, and facilities. The bottleneck for the data flow currently is real-time data transfer, measured in minute and second time increments and stored on a real-time server, to the engineers' desktops in a clean and timely fashion (Stundner & Oberwinkler, 2004). At Eni the historian database is the first point of storage of the field data, the high frequency data is processed through a quality-control system, stored and aggregated when required in CPOD (Fig.…”

Managing LNG Deliverability: An Innovative Approach Using Neural Network and Proxy Modeling for Australian CSG Assets

“…SOM clusters data such that the statistical relationship between multidimensional data is converted into a much lower dimensional latent space that preserves the geometrical relationship among the data points (Roy, Marfurt, & Castro, 2010). Unlike conventional deterministic problem-solving algorithms, Neural Networks can be trained to perform a particular task based on data /data-driven information (Stundner & Oberwinkler, 2004). SOM produce a map of usually one or two dimensions and plot the similarities of the data by grouping similar data items together.…”