Case Study: Reservoir & Well Diagnostics with ML
Client Challenge: Mature onshore oil field with declining rates and rising water cut. Conventional well tests were slow, inconsistent, and inconclusive and there was uncertainty on reservoir connectivity and well performance drivers
Solution: Integrated production, well test, and surveillance data. Applied ML clustering to group wells by performance patterns.Used anomaly detection to identify wellbore damage & flow restrictions. Built ML-assisted reservoir connectivity map using tracer & pressure data.
Key Insights: Early water breakthrough in 3 producers linked to high-perm streaks. 2 wells identified with completion damage restricting inflow. Undeveloped southern zone with missed recovery potential.
Impact Delivered: +8% forecasted recovery factor over 5 years. 60% faster diagnostics vs. conventional workflows. Reduced intervention costs with targeted workovers & restimulation. Data-driven roadmap for field-wide production optimization.
More Case Studies Available on request
| Aspect | Machine Learning | Deep Learning | Statistical Methods | Examples | |
|---|---|---|---|---|---|
| 1. | Core Concept | Uses algorithms to learn patterns from data and make predictions or classifications | Uses multi-layer neural networks to automatically extract complex features from large datasets | Relies on mathematical modeling and probability theory to infer relationships and test hypotheses | Predictive maintenance, quality control, performance optimization |
| 2. | Data Requirements | Works well with moderate amounts of structured data | Requires large amounts of labeled or unlabeled data for training | Can work effectively with small to moderate datasets if assumptions are met | Engineering simulations, sensor data analysis, lab experiments |
| 3. | Model Examples | Decision trees, random forests, support vector machines, k-means clustering | Convolutional neural networks (CNNs), recurrent neural networks (RNNs), transformers | Regression (linear, logistic), ANOVA, hypothesis testing, time series models | Structural analysis, fault detection, energy consumption forecasting |
| 4. | Advantages & Limitations | Good balance between interpretability and accuracy; may struggle with very complex data | Excellent for complex, high-dimensional, or unstructured data; requires high computation and is less interpretable | Highly interpretable and theoretically sound; assumes data distributions and may oversimplify relationships | Trade-off between accuracy, interpretability, and computational cost in engineering problem-solving |