One of the most difficult challenges of digital oilfield projects is requiring people who would not choose to work together, to collaborate for success. Operators and Data Scientists tend to be very different people, and these types of projects push them to work together, typically under fire from management to deliver results. Projects can fail simply due to the failure of these groups to get along. Collaboration is the key to getting success in this situation.
How do you foster cooperation between groups with such different skills and backgrounds? Operators and Engineers have years of experience and a deep understanding of petroleum engineering in the real world, but they are busy in the field striving to get the most out of their wells.
Operators think data science cannot understand the complexity of their operation. Data science thinks models can be built from data and operators will use them. Management thinks they can give data to data science and operational improvements will ensue. Failure of many projects is due to the failure of these groups to understand one another and collaborate to achieve shared objectives.
All three parties are mistaken in what is needed and what will lead to success for this team. Expert Guided Machine Learning assumes these groups need to be encouraged to work together. It provides a shared environment so separate teams from different locations can work together towards a common objective sharing the same information.
Operators, engineers and subject matter experts (SME’s) are engaged early as they define the problems and solutions they seek. They create labels to train the model. They set the target. The collaboration environment allows all parties to share information and exchange ideas. Well and model performance can be reviewed through a common lens with each party reviewing and commenting on different events, labels, and tags.
Data Science gains an understanding of the complexity of well operation as they learn from operators. Everyone can review the results of models in training and production along with underlying performance statistics like accuracy, precision, recall and the associated features and characteristics that trigger the models.
Operators learn the importance of labels as they see how they define what the models are looking for.
Models can be validated efficiently by observing well performance. If a gas lock model warning ends in a shutdown, it can be automatically validated. If a recommended change in speed avoids the shutdown, then that can be identified automatically as well. We call this auto-validation. Overall performance improvements can be tracked for management to see results in the context of field operation, production, and profit & loss.
The validated results of the models subsequently create more labels, so labeling is iterative with the effort of labeling reduced with each iteration.
To learn more about the collaboration of operators, managers, and data science into the creation of AI/ML models, access our white paper entitled “Expert-Guided Machine Learning: Engaging Petroleum Experts’ Know-How in Petroleum Engineering and Oilfield Operations” here.