Leveraging Probabilisitic Multivariate Clustering Analyses of Well Logs to Identify “Sweet Spot” Intervals in Heterogeneous Conventional and Unconventional Reservoirs
- Eric Eslinger (eGAMLS Inc.) | Francis Boyle (eGAMLS Inc.) | Alan A. Curtis (eGAMLS Inc.)
- Document ID
- Society of Petrophysicists and Well-Log Analysts
- SPWLA 60th Annual Logging Symposium, 15-19 June, The Woodlands, Texas, USA
- Publication Date
- Document Type
- Conference Paper
- 2019. held jointly by the Society of Petrophysicists and Well Log Analysts (SPWLA) and the submitting authors
- 0 in the last 30 days
- 120 since 2007
- Show more detail
The results of a multi-well Bayesian-based Probabilistic Multivariate Clustering Analysis (PMVCA) with a built-in prediction subroutine can be leveraged to rapidly identify “sweet spot” Intervals in well profiles that satisfy a set of user-defined cutoff criteria. Usually, the primary goals are to generate total porosity (Por) and total water saturation (Sw) profiles within the Intervals. The method works for both conventional and nonconventional reservoirs. The required input data is triplecombo well logs, although more credible results can often be obtained when whole core plug data such as grain density (GD) and Por are available. The work flow begins with a PMVCA using well logs (after well log QC, including curve normalization, and rigorous core-to-log depth correction when core data is available).
The thickness plus the average and standard deviation of Por of each Interval are determined using a modified density-porosity (DPHI) equation. A modified DPHI equation is one in which GD can vary with each sample (digitized depth step). In the absence of any core data for calibration, GDs might be estimated for each cluster (Electroclass) or for each generic Lithology (into which the Electroclasses have been assigned) using an equation that considers the probabilistic assignments of each sample to each Lithology. An alternative approach for determining GD exists if at least one well involved in the multi-well PMVCA has sufficient and credible plug GD analyses. Then, a complete GD profile can be generated for not only the cored well, but also each non-cored well if GD from the cored well is included as a clustering variable in the PMVCA.
In an analogous manner, Sw profiles within each Interval can be generated using a modified Archie equation, where “modified” means that the a, m, and n constants can vary with sample in the same probabilistic manner if different a, m, and n values are assigned to different Electroclasses. In conventional reservoirs where wet zones might exist, a Pickett plot linked to the PMVCA Electroclass results can assist with determination of Rw and Archie m for different but closely associated Electroclasses.
Crucial to the use of the PMVCA probabilities in the computation of Por and Sw is the development of a realistic PMVCA. Two options exist for initialization of a PMVCA: supervised and unsupervised. Due to data and logistics constraints, unsupervised initialization is usually used in unconventional plays. With this method, training can be done using all valid data when using a model-based clustering engine that has a neural aspect. The use of multiple runs (realizations) combined with selected convergence criteria assist with identifying an optimum solution.
|File Size||2 MB||Number of Pages||16|