Gaining Value & Control for Operators with Expert-Guided Machine Learning
Once validated models are created and they match what operators want, the final step is to deploy the models on live data. This is when we actually deliver value in real-time to help operators address problems and make changes quickly. All too often, model training and deployment are two very different steps. The data and transformations that are used to build the model typically need to be implemented onto completely different systems. These systems, such as SCADA, well management and ticketing systems, are not designed for this and the effort to re-engineer these systems can seem overwhelming.
A Common Environment for Training and Deployment integrates and interoperates with these other systems. Sensor, well and operational information appear in context with the warnings and recommendations from models. This environment doesn’t replace the other systems but augments them by tapping into the data streams that feed the models and then sending notifications to the operators in the field on their current ticketing systems.
The transition from training to a product should be seamless and immediate. And the models should produce results in production that are consistent with the results in training.
Even when models are deployed, they can still fail to get buy-in. If operators don’t take action, nothing happens with the models. Engaging operators early helps, but most of the interaction between operators and models are in production while wells are operating. To get necessary adoption and action on recommendations, operators need to understand the trends and rationale behind them. Visualizations that highlight the timing of changes in key signals, changes in shapes of dynacards, performance trends, and other underlying indicators can assist in gaining confidence in the models.
Operators must also interact with notifications and provide feedback indicating when the models are right or wrong and identifying events the model missed. Their feedback flows back into the model improving future performance. With more training, the model responds to their feedback. They can increase or decrease thresholds to get more or fewer notifications. They can change the severity of the message. They can even tradeoff between nuisance notifications and potentially missing a critical event. This builds support in operations as they realize they are in control and learn to use these models to improve their own success.
Essentially, we are enabling operators to collaborate not just with data scientists, but with the models themselves.
Expert Guided Machine Learning closes the loop. The loop from labels to training and running models to taking action and providing feedback that in turn creates more labels. And the loop from operators to data scientists and back to operators.
Expert Guided Machine Learning is not different than data science or AI. It simply makes data science and AI work successfully in the digital oilfield. It is a combination of process and environment that enables collaboration making supervised learning efficient and effective. OspreyData provides that combination.
The OspreyData solutions integrate with your systems: data collection, SCADA, well management, work order/ticketing, and other operational systems. It collects the data required and then orchestrates machine learning to train the underlying AI models that improve well performance.
Our clients, their operators, and engineers have used the solution to collaborate with our data scientists and SME’s to build models for thousands of wells across major fields and covering most major lift types. We understand how to collaborate with your teams to understand your operation, work together to develop models that address your challenges and take action to take advantage of these innovations.
If any of you are interested in trying out the solutions that we offer, we welcome you to contact our sales team to learn more about these offerings and to discover the benefits of our Expert Guided Machine Learning solution firsthand.