RESEARCH POSTER AWARD FINALIST: HADEER: Hybrid AI Driven Engine for Enhanced Reservoirs

This research poster award finalist presentation introduces HADEER, a hybrid AI-driven engine for enhanced reservoir management in subsurface energy, presented in a session chaired by Tanzima Islam (Texas State University). The work targets two bottlenecks that force uncertainty quantification and closed-loop control offline: seismic and well-log data too large and heterogeneous to move and regrid repeatedly, and full-physics reservoir simulations that cost days of CPU time. The guiding principle is to learn representations once and reuse them. For data, a neural field represents the seismic volume as a continuous function queryable at any coordinate and resolution; on the F3 offshore survey it reconstructs the volume at 30 decibels of fidelity with up to 136x compression while fusing sparse well logs with dense seismic data, and its weight space encodes geological similarity. For physics, a physics-aware AI proxy trained on 500 examples of the SPE9 community benchmark reproduces field-scale production within 11-18% of the full simulator while running 240 times faster, turning a thousand-scenario sweep from eight hours into two minutes. A reinforcement learning agent then uses the proxy for closed-loop pump control, discovering operating regimes yielding 16% more cumulative oil production with over 15% less water, with per-pump control more than doubling gains over field-wide control. The authors present the approach as a general blueprint for AI-accelerated simulation science beyond oil and gas.