The processed and analyzed production data is leveraged by our machine learning system, which orchestrates multiple data-driven models working in parallel and together. We combine hierarchical neural networks with first principle physics to learn the real dynamics, correlations and uncertainties in the production system. In operation, as new data becomes available, online learning ensures that the models stay calibrated, removing the need for large model maintenance efforts.
The simple interface of the data-driven models allows for a connection to conventional flow simulators, to form a hybrid-AI. Virtual data points from the flow simulator can reinforce the predictive capabilities of the hybrid-AI.
This opens up for a broad spectrum of production enhancing applications of varying complexity.
The AI´s learnings are made available to the production team to both use directly and improve their situational awareness, whether to assess the quality of production settings or to analyze well tests in real time. Some example applications include live assessment of well response to control changes, flagging of underperforming wells, measuring uncertainty and tracking progress of well deduction tests, and monitoring and predicting slugging.
The production team can capture significant value through avoiding production losses and improving its operating policies already by applying ProductionCompass AI.