Tunes the reservoir model until it reproduces what the field actually produced, automatically and with the uncertainty kept rather than discarded.
Manual history matching takes three to six months per field and produces a single "matched" model whose uncertainty has been argued away rather than quantified. Decisions are then made as if that one model were the truth.
An ensemble smoother with multiple data assimilation updates the property ensemble against production history, keeping a family of matched models rather than collapsing to one. Parameter groups are constrained to geologically plausible ranges agreed with the geomodeller before the run.
The method assimilates the data it is given, so allocation error in the production history becomes a property of the matched model. In fields with back-allocated rates and no well-test discipline the result is a model that matches wrong numbers precisely. We run a data-quality gate first and have refused this work where the gate failed.
Cada página de capacidad declara una limitación real. Si un proveedor no puede decirle dónde termina su producto, no le están diciendo la verdad sobre dónde empieza.
A learned stand-in for the full reservoir simulator that answers in seconds instead of hours, so an engineer can test fifty scenarios instead of three.
Produces P10, P50 and P90 production forecasts whose spread is calculated, not chosen to look reasonable.
Continuously checks incoming data for stuck sensors, impossible values, missing periods and broken units, and reports the field’s data health as a number.
Works out which injector is actually supporting which producer, and where the injected water is being wasted.
Fills the rock between wells with geologically plausible detail instead of a smooth interpolation that no outcrop ever looked like.