Practical use of Neural networks in Tight Gas Fractured Reservoirs: Application to the San Juan Basin

Abstract An innovative approach that uses neural network technology is developed to identify highly-productive "sweetspots" in the naturally fractured upper Dakota Formation, San Juan Basin, northwest New Mexico. Dakota estimated ultimate recoveries (EUR) are predicted through the vertical and horizontal integration of thirteen reservoir properties taken from three horizons over a 24-township area. The vertical information includes pay thickness, and resistivity (geological facies) from three gross sand units that contribute to the comingled upper Dakota production. The horizontal information includes, thickness, and resistivity maps for each sand unit as well as structural property maps over the entire 24 townships area. Using a neural network, relationships are established between the thirteen geologic parameters and fracture intensity inferred by the EUR. These geologic models indicate that the parameters that control upper Dakota production are in order of relative importance: thickness, resistivity and structure. The derived relationships are used to predict "sweetspots at locations without well control." P. 573