Supervised Learning Workflow for Predicting Reservoir Properties in AttributeStudio 8.3
Date: Thursday, November 7, 2019
Time: 9:00 AM-10:00 AM (MST)
Click here to register
The new deep-learning toolbox in AttributeStudio 8.3 provides the best-in-class algorithms, yet a very easy-to-use workflow for supervised learning between well data and seismic attributes. In this webinar, we will use a real data set to demonstrate how to use supervised learning network for predicting both discrete and continuous types of well data, such as facies, NTG, and porosity in a reservoir interval.
A supervised learning workflow in AttributeStudio consists of the following steps:
- Determine the interpretation object on which to predict reservoir properties: a horizon surface, an interval or a strata-grid.
- Select wells to define a training object
- Upscale the target well property, such as facies log or porosity log, to the horizon, interval or strata-grid
- Extract seismic attributes to the training object
- Train the supervised learning network between the target well property and a selected set of seismic attributes
- Performance analysis of the supervised learning network resulting from step 5
- Apply the supervised learning network to predict the well property when the prediction performance is satisfied, otherwise go back to step 5
We will pay a special attention to the performance analysis function in training the supervised learning network, such as how to avoid over-training by selecting proper data for training, validation and blind testing.