No abstract

No abstract

No abstract

ACKNOWLEDGMENTSWe thank Chevron Petroleum Technology Company for the permission to publish this material. We also thank the following colleagues at Chevron: K. L.Finger and S. R. Jacobson for supplying the biostratigraphic and geochemical interpretations used in this study; N. 1. Hancock for aiding the interpretation of lithologies from well logs; and D. S. McCormick for his critical review and comments of the manuscript. ABSTRACTFluid-flow simulation results are used extensively as reservoir performance predictions upon which to base economics for reservoir management decisions. The generation of numerical models for simulation purposes may be easily facilitated by computer-aided algorithms, regardless of the quality of data or input parameters, This study focuses on potentially erroneous conclusions deduced from simulator predictions if the simulation models are built without a sound geologic framework and the incorporation of seismic-based information, Models derived from two different correlation strategies, lithostratigraphic and chronostratigraphic, using well logs only are compared. Seismic inversions are included within the chronostratigraphic framework as a third model type. Histograms of breakthrough times indicate little difference between the well only models of either correlation strategy, whereas water displacement pattems are significantly different. Selection of the appropriate simulation results is closely tied to the reservoir management objective in question. Models which have been conditioned by the seismic pseudo-logs show substantially different results for both breakthrough and displacement distributions, and the spread or uncertainty in breakthrough estimates is greatly reduced compared to the log-only model results.A cloud transform method with correlated probability fields is introduced for stochastically estimating one model parameter (porosity) from another (impedance). This method allows incorporation of the scatter in the relationship between the two parameters (crossplot).References and illustrations at end of paper, 601

A reservoir model was generated to quantify the enhancements in reservoir description resulting from the incorporation of seismically derived information with well-based data and to study how the quality of the seismic data affects those reservoir descriptions. The reservoir model was created from 56 well logs with a predetermined spatial correlation using geostatistical techniques. A regular grid of traces was extracted from this model and transformed into 3D seismic surveys which were then inverted to produce acoustic impedance logs. Several seismic data parameters such as frequency bandwidth, waveform phase estimation error, and correlation coefficient between seismic and well data were varied to study their relative importance on reservoir description and subsequent flow results. These differing seismic data sets were finally integrated with the original well-log data again using geostatistical methods. Comparative fluid displacements were performed on all reservoir realizations using an all-purpose flow simulator to quantify the flow behavior differences. The results. based on breakthrough time and oil recovery predictions, show that even marginal increases in frequency bandwidth (e.g., increases of 10 Hz in the upper frequency) provide significant enhancements to the reservoir model. Also, estimation of waveform phase to within +/- 20 degrees of the true value is shown to be sufficient to capture much of the interwell heterogeneity contributed by the seismic data and thus does not significantly affect fluid-flow results. Finally, even for correlation coefficients as low as 0.4, which initially was considered quite low. sufficient positive contribution from the seismic data still warrants integrating it with the well-log data. Introduction In recent years, recognition of the important impact reservoir heterogeneity has on fluid displacement processes has led to the use of detailed, geologically realistic, quantitative reservoir models as input to fluid flow simulators. Such large models require the assignment of petrophysical and flow properties to each location represented in the numerical fluid flow simulation. Unfortunately, for most reservoirs, such data are only available at a few discrete sample locations, namely wells. Property values can be assigned to the remaining grid cells by interpolating the available well data or by generating stochastic realizations, each of which is a possible representation of the reservoir given the available data. Both of these methods have serious drawbacks. Smooth interpolations of properties are known to produce biased recovery predictions in fluid displacement calculations. On the other hand, stochastic realizations conditioned by very few data tend to generate recovery curves that cover a very large range of values, as will be seen later in the results section. Such problems may be alleviated by integrating information from 2D and 3D seismic data, which are available on a much denser areal coverage than wells, extending the use of such data beyond its traditional application in mapping large-scale subsurface structures. Seismic trace inversion to pseudo-logs provides acoustic data which can be used as soft information which, in turn, can then be used to infer the variation of petrophysical properties in the interwell region. Seismic data are considered soft because they differ in several very important aspects from well-log data and core measurements. First, the vertical resolution of the two types of data is vastly different. Well-log data provide, for all practical purposes, a point value every fifth of a meter while seismic data gives a rock property averaged over a very large volume, usually on the order of tens of cubic meters. The size of that volume is determined by the frequency bandwidth of the recorded seismic signal and the type of rock being traversed by the signal. P. 913^