Fourier neural operator-based temperature field prediction model for fractured geothermal reservoirs: addressing diverse fracture morphologies and injection-production parameters

Geothermal energy is a pristine source of clean energy. The repetitive numerical simulation of diverse injection parameters and fracture morphology represents a pivotal aspect in optimizing the development strategy for the efficient exploitation of fractured thermal reservoirs related to a number of practical demands: (1) quantifying how fracture morphology patterns govern thermal migration and heat extraction efficiency (2) simulating injection protocols to mitigate premature thermal breakthrough, and (3) assessing long-term reservoir sustainability under variable operational loads. Considering conventional numerical simulations struggle with computational complexity and latency, we propose a dataset construction framework coupled with a Fourier Neural Operator (FNO) model tailored to capture the interplay between fracture morphology variability and boundary condition dynamics. Initializing with near-pristine temperature fields, accurately mapping temporal temperature evolutions across 6-, 24-, and 60-month afterwards while accounting for geometrically complex fracture networks and operational parameters such as injection temperature/flow rate. Integrating these specialized models, we are able to generate extensive prediction results over an ultra-long period of time, spanning from 5 to 15 years, using a relatively arbitrary input within 1 min, requiring one input. Resolving these critical engineering unknowns with high computational efficiency, the framework enables real-time adaptive workflows, risk-informed drilling decisions, and sustainable yield maximization—advancements for geothermal project viability in fractured reservoirs.