Ranking waterflood candidates with machine learning: a case study for Pan American Energy.
For Pan American Energy we built a machine-learning methodology that ranks candidate areas for waterflooding and predicts their estimated ultimate recovery (EUR) with its uncertainty. It was delivered in seven months and won the Best Work Award in the Conventional Section at CONEXPLO 2026.
Problem
In a large waterflooding operation, converting areas from primary to secondary recovery is routine. Assessing and prioritizing candidate areas is time- and labour-intensive, and sometimes not feasible within the time frames operations require.
Approach
We developed a machine-learning methodology that ranks candidate areas and predicts their EUR (estimated ultimate recovery) with a stated uncertainty, trained on the company's own existing projects.
Key challenges
- Data integration: disparate sources required a robust pipeline and rigorous data QC.
- Ambiguous dataset: required careful algorithm selection and hyperparameter tuning.
- Geological and engineering complexity: petroleum-engineering principles were essential to build features with real predictive power.
Results
- Methodology delivered in 7 months, including a statistical model to estimate missing permeability values.
- Validated against known primary and secondary production results, showing strong consistency with the expert judgment of the company's SMEs.
- Applied to candidate areas, producing a ranking of areas and reservoir layers with estimated EURs and their uncertainties.
- Full characterization effort is now concentrated on the best areas, meeting the deadlines set by the pace of operations.
The work was presented at CONEXPLO 2026, where it won the award for best work in the Conventional Section.
Facing a similar decision?
Tell us about your data and the decision it should support. The first call is free.