Over a terabyte of sensor data was harnessed to identify and predict well freeze events on 1,700 wells. The data was mostly from wellhead and separator tanks and included temperatures, pressures, tank levels, and flow rates. Also, operational reports of freezes and records of well design, well environment, and production history played predictive roles in the models. The data was large, diverse, noisy, often erroneous, and incomplete. With input from the client, key ratios that affect each well’s propensity to freeze within the next three months were calculated. PS AI Labs used Devron's federated learning platform to build models across the datasets that could predict freezing times better than any previous efforts. During the data discovery and model build process, the insights about predictive relationships were reviewed periodically with client subject matter experts, and proposed models were thoroughly vetted by the client using a blind validation process.

‍

With PS AI Labs' expertise and Devron's federated learning platform, our predictions for optimizing the timing of plunger pump installation was projected to reduce production deferrals by 15 to 20%. The Downhole Freeze model had a 70% success rate. The insights therein helped the client prioritize well interventions and optimize maintenance resources to reduce freeze remediation costs by millions of dollars. It was projected to save 600,000 MSCF (million standard cubic feet) in deferred gas volume annually.

‍