Geothermal doublet simulator: what will this doublet deliver over 30 years?
Place an injector and a producer, set the rates, and the lab simulates flow and heat right here in your browser: produced temperature, heat power, injector pressure and when cold water breaks through. Every answer is a P10–P90 from an ensemble of rock models.
The bed comes from the Plug-to-Block Lab, the rock between the wells from the Heterogeneity Lab. The Reservoir Lab takes both to the whole reservoir and runs it forward in time.
KAI-Pilot is the Reservoir Lab's copilot. Ask whether a doublet can deliver a rate at a temperature for so many years: it picks the data, runs or reuses the KAI engine ensemble, explains the result and drafts the study report. It may not invent a number: every figure links to the run and file that produced it, and an answer that cannot be traced is not released.
Answers here are on the Utah FORGE case, from stored KAI engine runs (simulated results). Signed in, KAI-Pilot answers your own questions in your workspace; a KAI engineer reviews the report before it is final.
What the preview solves, and what it leaves out
Solved: single-phase water flow between wells held at a mass rate, and heat carried by the water and conducted through the rock, implicit in time, on a small grid.
Ensemble: each member draws a different correlated permeability field; the bands are the P10–P90 of the members.
Left out in the preview: water properties that change with temperature, gravity, fractures as separate paths, and stress. The full run on the KAI engine adds the first two and the well controls the engine uses; the FORGE case adds fractures.
Checked: on the same plan with uniform rock, the preview breaks through 17–21 % earlier than the KAI engine and ends 6–11 °C cooler after 30 years (tools/test_reslab.mjs). Read the preview as a first look; plan on the engine run.
How the engine is checked
Against exact answers: Lauwerier's heat exchange with the rock, the doublet breakthrough of Gringarten and Sauty, Theis and Thiem for pressure.
Against open-DARTS on a 3D doublet: breakthrough at 4.906 against 4.924 years (0.4 %).
Upscaling on the SPE10 benchmark: a coarse well pair within 0.12 % of the fine model's rate.
On real data: the Utah FORGE wells, logs and 2024 circulation tests, with the misfits shown.