Tight gas reservoirs are regarded as one of the main unconventional hydrocarbon resources. Extensive drilling activity is in progress in North America, particularly in the tight gas fields of the Rocky Mountains. The common practice employs drilling multiple wells from a common pad site. Based on formation evaluation from openhole or cased hole log data, multiple stacked sands are completed through staged hydraulic fracturing operations. Maintaining high production levels requires continuous development drilling and fracturing programs that follow very tight schedules.
While openhole data may be preferred, the typically poor borehole conditions encountered in tight gas reservoirs make openhole logging impractical or impossible. Delays due to openhole logging problems can have a significant negative impact on the cost of drilling, logging and completing the well. As a result, the industry has shifted to rigless cased hole logging with pulsed neutron technology and measurements. While new cased hole gas saturation methods have been developed in recent years, one of the first uses of pulsed neutron logging in tight gas reservoirs involved development of neural network or multidimensional histogram models to emulate openhole log data based on the pulsed neutron responses. The emulated openhole data are then analyzed using conventional saturation methods. This approach is regularly employed using combined openhole and cased hole data from one well and applying the emulation model to predict openhole responses in multiple wells on the pad using only cased hole logs.
In this paper we present an innovative method to emulate basic openhole log responses (bulk density, neutron porosity and deep resistivity) for tight gas reservoirs. The emulation technique uses a feed-forward, backpropagation based neural network model and three-detector based pulsed neutron data. The three-detector pulsed neutron instrument provides greater formation sensitivity than conventional two-detector instruments and provides higher accuracy of emulated openhole data.
Results are presented from a study of three wells for which openhole and cased hole logging were performed. The development of the neural network model is discussed, and direct comparisons of the actual and emulated openhole data demonstrate the improved accuracy of the three-detector pulsed neutron emulation method.
URTeC 1596010