The problem of predicting the geometry of the multi-stage multi-fracture horizontal wells is an extremely complex task as shown in the images below:
Our workflow for model development is shown below:
- Collect field data for training (hydraulic fracturing parameters and microseismic recording).
- Find a relationship between the inputs HF parameters and output cloud of MSE. We used machine learning to find the relationship.
- Do a Monte Carlo simulation and use the Sobol Technique to represent the relationships function as summation of polynomials.
The advantage of our approach is a model that runs in a few seconds (compared to high fidelity models that have CPU run time up to days ), represents the SRV as a polynomial combination of input parameters in hydraulic fracturing process, and has a controlled error.
In the figures below, the overall workflow of the FACT’s team development is presented from left to right.
- Table containing the hydraulic fracturing parameters
- ANN model predicting the SRV as a result of each fracturing step.
- Sobol technique to represent the relationship with simple polynomials.