DATA MINING AT THE NEBRASKA OIL & GAS COMMISSION

The purpose of this study of the hearing records is to identify factors that are likely to impact the performance of a waterflood in the Nebraska panhandle. The records consisted of 140 cases. Most of the hearings were held prior to 1980. Many of the records were incomplete, and data believed to be key to estimating waterflood performance such as Dykstra-Parson permeability distribution or relative permeability were absent. New techniques were applied to analyze the sparse, incomplete dataset. When information is available, but not clearly understood, new computational intelligence tools can decipher correlations in the dataset. Fuzzy ranking and neural networks were the tools used to estimate secondary recovery from the Cliff Farms Unit. The hearing records include 30 descriptive entries that could influence the success or failure of a waterflood. Success or failure is defined by the ratio of secondary to primary oil recovery (S/P). Primary recovery is defined as cumulative oil produced at the time of the hearing and secondary recovery is defined as the oil produced since the hearing date. Fuzzy ranking was used to prioritize the relevance of 6 parameters on the outcome of the proposed waterflood. The 6 parameters were universally available in 44 of the case hearings. These 44 cases serve as the database used to correlate the following 6 inputs with the respective S/P. (1) Cumulative Water oil ratio, bbl/bbl; (2) Cumulative Gas oil ratio, mcf/bbl; (3) Unit area, acres; (4) Average Porosity, %; (5) Average Permeability, md; (6) Initial bottom hole pressure, psi. A 6-3-1 architecture describes the neural network used to develop a correlation between the 6 input parameters and their respective S/P. The network trained to a 85% correlation coefficient. The predicted Cliff Farms Unit S/P is 0.315 or secondary recovery is expected to be 102,700 bbl.