Uncertain Models - Learning from the past and optimizing for the future
summary
Shows how Equinor turns uncertain reservoir and windfarm models into actionable plans by sampling ensembles, running each concrete model through existing simulators, and applying ensemble Kalman filter / ensemble smoother updates to fit production, seismic and gravity data. Demonstrates an optimizer that computes gradients by assigning perturbations across ensemble members to cut simulation count, gives a toy example (train at 5 m/s, naive position 24.37 vs filtered 25.35 after five measurements) and ships working software—open source ERT—used by Norsk Regnesentral, TNO and Norce.
AI-written summary · about
chapters (9)
- 0:00Start
- 3:56A toy example
- 8:26Kalman filters
- 10:22Playing around with the equations
- 13:19Applying the Kalman filter
- 26:29Implementing uncertainty
- 31:19Executing the plan
- 32:43What do we observe?
- 34:40Summary
related talks
01h 10mBayesian Optimization and Machine Learning for Accelerating Experiments in the Physical Sciences
37mFast-Forward(ing) Models with AI Emulators
33mPhysics-Informed Machine Learning for Real-time Reservoir Management
01h 11mHidden Physics Models: Machine Learning of Nonlinear Partial Differential Equations
34mData-Driven Solutions in Engineering
31mML time series forecasting methods
26mAddressing Some of Challenges When Optimizing Long-to-Train-Models
28mInterpretable and realistic generative models in data science?