Industrial ML
Pump Degradation and Flow Rate Estimation
Estimating electric submersible pump (ESP) degradation and liquid flow rate from operating data across hundreds of wells.
Modeling pipeline
- Data preparation, feature engineering, and feature selection.
- CatBoost models with cross-validation and Optuna tuning.
- Experiment tracking in MLflow.
Evaluation
Models were developed using operating data from several hundred wells. Evaluation used later periods from the same wells, with earlier periods used for training.
MacroMAE is the mean of the MAE calculated separately for each well, giving each well equal weight.
Degradation coefficient
Including gas and viscosity effects.
| MacroMAE | SMAPE | R² | Target range |
|---|---|---|---|
| 0.042 | 4.37% | 0.85 | 0–3 |
Liquid flow rate
| MacroMAE | SMAPE | R² | Target range |
|---|---|---|---|
| 13.8 m³/day | 7.93% | 0.95 | ~0–1,500 m³/day |
Mechanical degradation coefficient
| MacroMAE | SMAPE | R² | Target range |
|---|---|---|---|
| 0.053 | 5.38% | 0.77 | 0–3 |