Have you ever struggled with forecasting young PDP wells? Geology, completions design, and well spacing are critical predictors for well performance, and by incorporating them into early-time forecasting, you can dramatically improve accuracy. Here, I'm showing results from our PDP forecasting models on younger wells from the Permian, Williston, and Appalachia. These models use machine learning to directly forecast declines, incorporating not just prior production but also geology, completions, spacing, parent-child, etc. This data is especially useful for younger wells, where the production data may be too noisy or non-hyperbolic to get a good fit using a traditional Arps curve. This is especially the case with public data sources! Across these three basins, our machine learning PDP forecasts improve accuracy by 32%, 46%, and 29% on the primary production stream, compared with a traditional DCA methodology. Why does this matter? Newer wells (<1.5 years old) make up about half of production in the L48, though it's even higher for some basins and assets. Poor forecasts on young wells leads to incorrect valuations, inaccurate supply models, and bad type curves. Our operator customers have been using these models for over three years now to improve their forecast accuracy, but we have recently released these forecasts as a live, auto-updating data product for our subscribers. At URTeC in two weeks, we are diving deeper into this technology as applied in the Permian Basin. Thanks to lead author Kiran Sathaye for his insightful work in this study. For more information, head to our URTeC page (link in the image). #oilgas #permian #urtec
Victor RONDON
8 个月
Forecasting with simpler tools like rapid miner
Reservoir Engineering Manager at Permian Resources
8 个月Invert the y axis. These look a lot like cum oil curves and the RE mind panics seeing Novi ML “well” 32% worse. With inversion you will look like a nice shallow decline well on rate-time plot to grab attention