Petrel Classification & Estimation
Handle Data Estimation & Forward Modeling Problems
Neural networks have emerged as proven technology to handle property estimation and forward modeling problems. The Classification and Estimation module provides an alternative to the current Petrel deterministic and stochastic 3D property estimation techniques, and introduces new workflows for log estimation, property mapping, and seismic facies classification. The module also introduces the new Trend Modeling process, which allows the estimation of 3D facies probability volumes based on upscaled well data and optional secondary variables such as seismic attributes.
Train estimation model process—how it works
This process gives you access to tools for neural network analysis, enabling you to train and then create the estimation model object. Input data types include
- well logs
- surfaces with attributes, including seismic attribute maps
- 2D or 3D seismic data
- 3D properties, both discrete and continuous points with attributes.
Predictive modeling
Once nonlinear functions have been created, the estimation model can be used for predictive modeling on a wide variety of data types via the appropriate Petrel process:
- make well logs—well logs
- multitrace attribute generation—seismic attribute cubes
- facies modeling—discrete property generation
- petrophysical modeling—continuous property generation
- make surface—surface attributes (including seismic attribute maps).
Trend modeling
A realistic facies model built from all available integrated data is an important step in a reservoir characterization workflow, so pay reservoirs, volumetric estimations, and cell connectivity are more accurately estimated. The Trend Modeling process allows you to combine facies log interpretation and 2D or 3D seismic attributes to create vertical proportion volumes that can be used later to drive the distribution of facies in the 3D model.
Advantages
The Train estimation model process and related predictive modeling methods bring a generalized neural network implementation for the estimation of well logs, surfaces, seismic volumes, and 3D property models. It is an alternative to geostatistics when a nonlinear relationship exists between a set of input data and a given output, or when there is no single variable or set of two variables that provides an adequate correlation.
With the Trend Modeling process, the geologist is able to reproduce realistic sedimentary environments, bridging the gap between 1D and 2D data, geological concepts, analogs, and 3D modeling world.
Request More Information