Petrophysical Well Log Analysis through Intelligent Methods

Objectives/Scope: Estimating reservoir petrophysical parameters such as porosity, permeability is vital to an in-situ hydrocarbon reserve evaluation. However, conventional laboratory measurements of core samples are extensively costly and time consuming. This work particularly concerns about applicability of different intelligent methods in reproducing horizontal porosity and permeability as main petrophysical parameters of reservoirs. Methods, Procedures, Process: To do so, information in terms of log depth, caliper, conductivity, sonic, natural gamma, density and neutron, water saturation, percent of shale volume, type of lithology as well as horizontal porosity and permeability are collected from well logging (from a total number of 19 exploratory wells) in an oil field in middle-east. Such preliminary design of input-output variables validates later through results from implementing neural networks to come up with selection of optimum number of input variables. Statistical preprocessing also carried out in order to understand the variational correlation of the collected dataset. Four different intelligent techniques of: (i) conventional artificial neural networks, (ii) artificial neural networks based on PCA transformation, (iii) statistical (bootstrapping) neural networks, and (iv) combined neural and fuzzy logic (neurofuzzy inference) networks trained, validated and tested in reproducing available measurements. Results, Observations, Conclusions: Results from implementing mentioned comparative techniques show higher capability of statistical neural network based on bootstrapping approach for our case study. Such network with 11 input parameters provided reliable performances of 93% and 77% respectively through training and validation phases of estimating permeability as desired output. For estimating porosity, results from a trained network with 9 input parameters made 75% and 71.5% correlation fit to the observation values in training and validation phases, respectively. Novel/Additive Information: Results from this study are expected to be applicable in: (i) dealing with incomplete well log dataset, (ii) estimating required petrophysical properties of reservoirs and (iii) decision making about measurements design.