THE APPLICATION OF ARTIFICIAL NEURAL NETWORKS WITH SMALL DATA SETS: AN EXAMPLE FOR ANALYSIS OF FRACTURE SPACING IN THE LISBURNE FORMATION, NORTHEASTERN ALASKA

KAVIANI, Danial¹, BUI, Thang D., JENSEN, Jerry L.¹, and HANKS, Catherine L.², (1) Petroleum Engineering, Texas A&M University, 3116 TAMU - 507 Richardson Building, College Station, TX 77843-3116, [email protected], (2) Geophysical Institute, Univeristy of Alasks Fairbanks, 903 Koyukuk Drive, Fairbanks, AK 99775-7320

Artificial neural networks (ANNs) have been widely used for prediction and classification problems. In particular, many methods for building ANN's have appeared in the last decade. One of the continuing important limitations of using ANNs, however, is their poor ability in analyzing small data sets because of overtraining. We propose to use the approach of radial basis functions to solve this problem and have applied this method to the analysis of fracture spacing in the Lisburne Formation.

Comparing our results with those from other ANN methods and multivariate statistical analysis, we find that the proposed method gives a substantially smaller error than the other methods. The errors in predicted fracture spacing for the Lisburne using the conventional ANN and statistical methods are about 50% larger than those obtained using the proposed method. By having a method which predicts fracture spacing more accurately, we were able to more reliably identify the effects of such factors as bed thickness, lithology, structural position, and degree of folding on the spacing.

In petroleum engineering and geosciences, there are many cases where the number of data is limited because of expense or logistical limitations, e.g., limited core, poor borehole conditions, or restricted logging suites. Thus, these methods should be attractive in many petroleum engineering contexts where complex, non-linear relationships need to be modeled using small datasets.