With Expert-Guided Machine Learning (EGML), even if models are deployed, they can still fail due to inadequate adoption and ineffective change management. If operators fail to take action, nothing happens. EGML helps operators build confidence in AI/ML models by involving them in all stages from training to production to ongoing review and feedback. We have discussed the importance of engaging operators early previously, but most of the interaction between operators and models are in production while wells are operating.
Operators need to receive the recommendations, warnings and other notifications quickly. They need to see why these recommendations were made. Cryptic messages from some automated systems will not build confidence. To get buy-in and take action on recommendations, operators need to understand the trends and rationale behind it. They need visualizations that highlight the timing of changes in key signals, the changes in shapes of downhole dynacards, the comparative performance trends and other underlying indicators depending on the type of lift, problem, and model.
If operators are not convinced by model results or have confirmed through testing that the model is not fully correct, they have a platform to give feedback so the model can learn from false-positive recommendations. Where the model missed a problem, operators can label the event and enter comments, tags, and upload work over reports to help improve model performance. No model is perfect. There is a tradeoff between precision (i.e. how often notifications are incorrect) and recall (i.e. how many problems the model might miss). This tradeoff can be adjusted using confidence thresholds depending on the cost and impact of each type of problem. Management and operations adjust thresholds to achieve performance targets ensuring alignment between the performance of the model and the objectives of the business. This builds support in operations as they realize they are in control and learn to use these models to improve their own success.
The final critical success factor that defines Expert Guided Machine Learning is this closed-loop back-and-forth collaboration between:
- Artificial intelligence and well operations
- AI/ML models and real-world problems
- Operators, engineers and data scientists
The feedback loop drives continuous improvement as models improve based on actual performance and expert guidance. Operators and engineers feel empowered and gain confidence as they see their input incorporated and their own performance improve. Experts and machines working together perform better than either could do on their own – the ultimate goal of EGML.
After modeling thousands of wells across different fields, OspreyData has learned how to address critical challenges in AI innovation. To learn more about expert-guided machine learning, access our white paper entitled “Expert-Guided Machine Learning: Engaging Petroleum Experts’ Know-How in Petroleum Engineering and Oilfield Operations” here.