Production forecasting for Venezuelan upstream assets

Why Venezuela first

Choosing a first market is a statement about which problems the software has to solve well. Venezuelan upstream data exercises the hard parts of production forecasting immediately: long-lived heavy-oil assets in the Orinoco Belt, mature waterflooded fields around Lake Maracaibo, and production histories interrupted by facility availability rather than by reservoir behaviour.

A forecasting tool that only works on clean, uninterrupted monthly production is not solving the problem those teams have. Building against Venezuelan histories first forces the awkward cases — a missing production report, a month of allocated volumes after a restart, a partial month with a low days-on count — to be first-class rather than exceptions patched in later.

That work is not Venezuela-specific in its result. Every mature basin produces the same categories of messy data.

The public demo portfolio

The demo uses a closed Venezuelan universe of 17 producing assets and areas, split between the two regions that dominate the country's production.

Faja Petrolífera del Orinoco

Assets and areas in the Orinoco Belt, shown east on the regional map: Ayacucho, Boyacá, Junín and Zuata area references.

Cuenca del Lago de Maracaibo

Assets and areas in the Maracaibo Basin, shown west: Bachaquero–Lagunillas–Ceuta, Intercampo Norte, Lagomedio, Tía Juana Lago, Lagocinco, Lagomar, Lagunillas Lago and Rosa Mediano area references.

Where the location data comes from

This matters more than it may appear, because a map of an upstream portfolio invites the assumption that it shows wells. It does not.

The positions shown are WGS84 centroids derived from public Ministry of Petroleum and PDVSA area and block references, drawn over Natural Earth public-domain regional geometry. They provide area-level context — which region an asset belongs to — and they are not wellhead coordinates. No proprietary well location data is used or implied.

The same discipline applies to the production history in the demo: it is an illustrative series used to demonstrate the workflow, and every published benchmark must name the public dataset and test window behind it.

Forecasting on interrupted production histories

Venezuelan monthly production series routinely contain gaps that have nothing to do with the reservoir. A decline model that silently interpolates them produces a confident curve fitted partly to facility downtime.

PronoHorizon keeps those events visible instead. Months with no production report stay marked as missing rather than being filled in, volumes allocated after a facility restart are marked as estimated rather than reported, and days-on counts stay attached so a partial month is not read as a rate collapse.

The practical consequence shows up in validation: when a holdout test shows a large miss, you can tell whether the decline model was wrong or whether the well was simply down for compressor work that month.

Usable beyond Venezuela

Nothing in the forecasting workflow is bound to a jurisdiction. The import step accepts CSV, Excel or public well data with arbitrary column names, the Modified Arps model is basin-agnostic, and the holdout validation method only needs a production history long enough to hide part of it.

Venezuela-first describes where the product is being proven, not where it can be used.