Date of Graduation

2004

Document Type

Thesis

Degree Type

MS

Committee Chair

Samuel Ameri

Abstract

A methodology to generate synthetic wireline logs is presented. Synthetic logs can help to analyze the reservoir properties in areas where the set of logs that are necessary, are absent or incomplete. The approach presented involves the use of Artificial Neural Networks, as the main tool, in conjunction with data obtained from conventional wireline logs. Implementation of this approach aims to reduce operation costs to companies. Development of the neural network model was completed using a General Regression Neural Network, and four wells that included gamma ray, density, neutron, and resistivity logs. Synthetic logs were generated through two different exercises. Exercise one involved four wells for development and training of the network. Subsequently verification was carried out using each of the wells that were used to train the network. The second exercise used three wells for training and development of the network. A fourth well that was not used during training and calibration, was selected for verification. Three combinations of inputs/outputs were chosen to train the network. In combination “A” the resistivity log was the output and density, gamma ray, and neutron logs, and the coordinates and depths (XYZ) the inputs. In combination “B” the density log was output and the resistivity, the gamma ray, and the neutron logs, and XYZ were the inputs, and in combination “C” the neutron log was the output and the resistivity, the gamma ray, and the density logs, and XYZ were the inputs. After development of the neural network model, synthetic logs with a reasonable degree of accuracy were generated. Results indicate that the best performance was obtained for combination “A” of inputs and outputs, then for combination “C”, and finally for combination “B”. In addition, it was determined that accuracy of synthetic logs is favored by interpolation of data. As an important conclusion, it was demonstrated that quality of the data plays a very important role in developing a neural network model. A recommendation for future works is to do a very careful quality control of the data before a neural network model is built. Conversely it was concluded that lithologic heterogeneities in the reservoir do not affect performance of a neural network model in generation of synthetic logs.

Download

Share

COinS