Petrophysical Parameters Estimation and Lithology Mapping for Enhanced Reservoir Characterization using Computational Techniques

The role of well-logs for reservoir properties calculation has been established quite well. However, often, only standard well-logs are available to run routine calculations of storage, saturation and flow. Advance information about reservoir mineralogy, lithology and therefore effective geomechanical properties are generally missing because the well-logs required (e.g., Shear, PEF, Lithoscanner) for successful estimation of these properties are not acquired on routine basis. Additionally, when available, these logs are limited in their coverage of subsurface sections. Drilling core plugs from the well or empirical correlations from analog settings are used to perform the reservoir engineering property analysis. Therefore estimation of the advance engineering information in reservoir, particularly in absence of hard controls (core data) and sparse coverage from logs is a challenging task. Additionally, such estimates and their geological significance is rarely verified and accounted for.

In this work, we present a comprehensive workflow for estimation of petro-physical and engineering properties, using empirical co-relations and its conformance through computational techniques such as ANN and other machine learning algorithms. Typically, the network is trained by some sample data with prior geological information to predict lithofacie and estimate other petro-physical parameters using empirical relationships and correlation based cross plots. We began in the under-represented carbonate reservoirs by predicting the missing and discontinuous well-log statistically (ANN, Gradient Boosting regression) and testify them by binding the predictions to lithology. We tackled the issue of core sample redundancy where the predicted shear log values and other suitable logs cluster to give lithofacie distribution which validates our physical model (multi-mineral solver, using convex optimization method). We rigorously examined the usability of limited number of logs in under-determined scenarios to predict formation mineralogy and porosity and suggested a set of combinations that will work better than other combination in a similar under-represented reservoir. A sensitivity check is also made by using different combinations of logs for determining the statistical and physical models.