NET PAY zone prediction in the gas field using Artificial Intelligence and Neural Networks
The real issue in the petroleum industry is that most wellbore measurement tools do not provide a complete understanding of reservoir properties and do not directly indicate the availability of hydrocarbons in place. As a result, extensive resources are required to establish further geological interpretation. To save resources and avoid human bias, we developed “Net Pay prediction using AI”.
The main goal of the research project is to predict petroleum net pay zones from well logging and core data. To do so, the net pay result was extracted from the production logging tool as a binary indicator (net, non-net) and then predicted using several AI techniques, including multilayer perceptrons and random forests. From the total of 10 wells with known net-pay zones, several well logs were used as input parameters, such as gamma ray, resistivity, sonic, density, and neutron porosity. Further, for the non-binary approach (oil, gas, and water), additional mud-gas data was used for a better match with the gas-only producing net-pay zones.
In the next stage, the core data from a well was included for small-scale predictive analysis. Hence, core mini-permeability measurements and CT-scan measurements, as well as the core bright-light and UV-light images, were analyzed and matched against each other to understand the relationship between the attenuation values and permeability parameters of the rock.
As a result of the blind test, using only well-log data as the input for the built Neural Network, net-pay zones from the PLT measurements were truly predicted, with a degree of match up to 85% for the binary approach and up to 79% for the non-binary approach. Analysis of the computer tomography scan of the core revealed lithological patterns that were impossible to see with the naked eye but had a high degree of match with the miniperm results.