|
|
GridSTAT is a unique and powerful tool for data quality control of well logs, core, markers, deviation survey, time depth relationship, seismic interpretations, and seismic data. Data problems can be identified and corrected or bad data removed relatively quickly. Thousands of well logs can be normalized in a relatively short time.
A reservoir model will not be better than the data that you feed into the model. A large part of the project time is often spent getting the correct data.
Geostatistics is a practical tool and integral part of GridSTAT.
By calculating Sw with resistivity and capillary pressure, water contact and free water level can be determined or verified. Water contact is for the determination of production without water, and is actually an average number. Between water contact and free water level, which may be as thick as a thousand feet or more, a well may produce water, oil, or gas in different proportions depending on the elevation, availability of oil, and porosity. Understanding and modeling of water contact and free water level is important for reserve estimation and production planning when large area is not covered by wells.
A geostatistical tool developed in GridSTAT provides help for correlation between wells. When well spacing is small, semi-automatic correlation may be used to provide initial marker. The the automatic correlation tool, it is possible to identify small faults that are otherwise difficult to identify.
Correct time depth convertion is critical for integrating seismic data with well data. Full scanning calibration developed in GridSTAT provides a unique tool for verifying the correct reference, adjusting the velocity profile for each well, and determining if the velocity distribution is reliable.
If the calibration window is small, the "match" may be faults. We suggest matching at least 1000 ms of well log synthetic to seismic, taking into consideration of geology and lateral continuity, using pseudo impedance as well as sonic log based impedance logs. By scanning a large time window with multiple logs (and multiple wells if available), we can determine how reliable the match is.
If a unique match cannot be achieved, the uncertainty may come from seismic acquisition quality, seismic processing quality, error in location of seismic or well (including deviation trojectory), error in the seismic time reference or well log depth reference, or quality of the well logs. In case of complex structure without prestack migration, the location of the seismic signal may be off and a reasonable search radius should be used.
Conventional seismic inversion is based on impedance. In practice, impedance log may not be in good quality or may not show good relationship to the reservoir attribute such as lithology or porosity. By selecting suitable pseudo impedance logs and extracting proper wavelets, seismic amplitude volume can be converted to lithology or porosity. Correct time depth convertion is essential for seismic inversion. Time depth calibration accuracy can be improved by using pseudo impedance logs as well as impedance log.
Let us consider a sandstone reservoir. If cross plot of impedance log with Vsh is examined for the different depth range or zones, it will become obvious that for some zones there is no clear relationship between impedance and sand. Even if a good impedance inversion can be obtained for these zones, it will be difficult to identify the sand bodies. The usual approach is a labor intensive and subjective process to interprete the sand bodies from impedance volume.
Our new approach is to use Vsh and effective porosity as pseudo impedance for time depth calibration and inversion, as well as using impedance itself. The time depth relationship will be the same for impedance and pseudo impedance, but the wavelet will be different for impedance, Vsh based pseudo impedance, and effective porosity based pseudo impedance. Vsh or effective porosity based inversion provides better definition of the sand bodies, and the work of interpreting sand bodies is now mostly based on a cutoff with minor interpretation. Effective porosity based inversion also provides softdata for building porosity model.
Suppose that a new well is drilled (or a well that was not used in the inversion), the Vsh log predicted from the impedance inversion probably would not show good correlation with the actual Vsh log. On the other hand, Vsh based inversion would show better prediction for Vsh, and effective porosity based inversion would show better prediction of the effective porosity, compared to impedance based inversion.
Seismic processing is often targeting structure interpretation. For attribute inversion, more high frequency energy should be preserved and relative amplitude should be preserved. In practice, reprocessing is costly, therefore filter and reverse-filter may need to be applied to prepare the seismic data for inversion. Since well logs and geological interpretations are usually available, this preparation should take well logs and geological information into account.
In order to define a reservoir with subtle structure and multiple water contacts, accurate structure model is critical. With limited well control, seismic interpretation with accurate time depth conversion becomes the key. Time depth conversion is based on a 3D velocity model from the surface down or from a known reference horizon down.
Information for building a 3D velocity model may include velocity profile at wells, seismic velocity, and layer velocity trends.
Velocity profile at wells can be the most accurate. Initial velocity profile may come from VSP or Checkshot. Sonic log may provide relative interval velocity. For each well with well logs, the initial velocity profile may be calibrated by correlating impedance or pseudo impedance based synthetic with seismic data. This can reduce the uncertainty in time depth conversion from around 50ms down to less than 10ms.
Seismic stacking or migration velocity is empirical, except in the case of prestack depth migration. With few wells, seismic velocity can provide information about lateral velocity variation. For best results, processing parameters should be checked so that the uncertainty of the seismic velocity can be estimated. Geostatistics may be used to integrated seismic velocity, which has larger uncertainty but wider lateral distribution, and velocity profile at wells, which has lower uncertainty but is areally limited, so that the final velocity model combines information from both sources with appropriate relative weighting.
Layer velocity trend is based on the assumption that velocity is a function of lithology and compaction. With a 3D earth model defining the earth as lithologically consistent layers, a velocity with depth trend for compaction can be assigned to each layer. Layers are determined geologically with help from seismic. Layer velocity is determined from well velocity profile and regional empirical data. Low velocity layers have large effect on the time depth model, and uncertainty can be large. Calibration with wells is critical for verification.
There are two main parts in the modeling of fractured reservoir. The first part is identification of fractured zones or effective fractured porosity and permeability at wells. The second part is prediction of fractured effects away from existing wells, with help from structure and seismic data.
Fractured effect at wells may be identified from well logs, mud logs, and drilling logs. Different reservoir may require different methods. Integrated evaluation is usually more effective. Drilling log of rate of penetration (ROP) is very helpful for fractured/caved reservoir description.
Depending on the origin of the fractures, different attributes may contribute to the prediction of fracture presence. These attributes may include lithology, elevation, structure and curvature, seismic amplitude, seismic frequency change, etc. Seismic data may be used in two different ways: one is inversion, and the other is statistical correlation, including neural network. For statistical correlation, there should be sufficient number of wells for the calibration to be statistically significant.
Prediction away from wells will be based on a combination of (1) seismic data, (2) structure, and (3) well data.
GridSTAT provides easy to use geostatistical modeling of reservoir heterogeneity. Upscaling of fine scale geological model is easy and practical. GridSTAT provides two options for modeling correct permeability heterogeneity. One is to build the model exactly the same dimension as reservoir simulation with the correct statistics. The other is to scale-back the statistics of the upscaled permeability model to that of the fine scale model. Both approaches have been proven to help reservoir simulation reaching reasonable history match in a short time.
Production may be predicted with the geological model using permeability thickness (K*H, Kg*H for gas, or Ko*H for oil) or oil/gas in place distribution (HCPV or HC H map).
GridSTAT loads production history data, draws bubble maps of production on top of geological prediction maps, and offers cross plot between production and geological model predictions for verification.
|
Send Email or Comments |