Smooth regression to estimate effective porosity using seismic attributes

Data mining is very important to characterize complex geological structures, where a large variety of geophysical and petrophysical variables are typically involved and interrelated. In this paper we apply smooth regression for data analysis, by means of the Gamma test (a revolutionary estimator of the noise in a data set) to aid in the construction of Artificial Neural Network (ANN) models to predict effective porosity (φe) using seismic attributes. As a result, we obtain the best combination of seismic attributes to estimate φe. We briefly describe the Gamma test, its benefits in model identification and model building. The first validation of the Neural Network based on leave-one-out was poor. Therefore, we generate a complementary set of synthetic data (from the original well-log data), varying the effective porosity and applying the Gassmann's equation for fluid substitution to obtain resulting velocities and densities. The complete procedure is repeated including the new synthetic well-logs and the best suited selection of seismic attributes is used to train a new ANN producing a better validation and more accurate results. The advantage of smooth regression over other techniques is that it tells us how well we can predict φe using any model. This information saves time during training of the ANN and also sets a lower bound for the mean squared error to prevent over-training.

► Estimation of effective porosity using non-linear modeling tools.
► Selection of seismic attributes via smooth regression technique
► Additional synthetic data generated using fluid substitution (Gassmann's equation)