03026nmc a2200229 4500 20260520010454.0 260224s xxu 068.82 553.28 C62 15730 Simposio de Desarrollo de Vaca Muerta MONITOREO ALTERNATIVO DE PRODUCCIÓN DE GAS CON TÉCNICAS DE MACHINE LEARNING 27/08/2024 ; 27/08/2024 14 p. ; 459-472 Fluid flow quantification is crucial for an accurate determination of individual well productivity. In blocks where production is mostly based on dry/wet gas, this information is obtained by dedicated flow quantity indicators (FQI) measurements. As an alternative to FQI measurements to estimate fluid flow, a common mathematical expression is the choke formula (CF). However, this formula is not always usable due to the assumption of a monophasic fluid flow (only gas). As a matter of fact, from own experience, for shale gas wells choke formula seems to perform typically very well in decline stages of production but poorly in ramp up stages where water production is not negligible. As part of monitoring procedures during the productive time of a well, many other variables are measured with an hourly frequency, such as: choke, pressures after and before choke, and temperatures after and before choke. This available information is part of all wells’ production history. The objective of the present work was to make use of other wells’ production history data for gas flow estimation in horizontal wells in order to avoid redundant FQI measurements and achieve cost reduction. For that, a Machine Learning (ML) based model was proposed, which was able to capture decline trends from historical production data. ML model’s performance was compared to CF estimations as a proxy for acceptable accuracy levels according to the Oil & Gas industry standards. The results obtained with ML model proved to estimate individual gas flow in a wide range of wells with an acceptable performance that, in 82% of total cases, was more accurate than choke formula based on R2 scores. In positive prediction cases, accumulated gas estimation with ML model showed a lower error than CF estimation, with errors being below the 5% tolerance margin. These results suggest an alternative way to exploit monitoring data to estimate gas flow production in gas fields. Achieved results demonstrate the high prediction potential of AI algorithms when combined with proper use of data. Further efforts may lead to a more robust model for a larger number of wells within the same fluid window. CONGTP Canatelli, Axel 61649 Dupuis, Godefroy 61814 Parigini, Lorenzo 61777 Congreso de Exploración y Desarrollo de Hidrocarburos (11er : 2022 nov. 8 - 11 : Mendoza) 61064 196672 196672 https://biblioteca.iapg.org.ar/ArchivosAdjuntos/Conexplor2022/Simp.VacaMuerta/vacamuerta26.pdf 0 0 0 0 264391 BVIAPG BVIAPG 068.82 553.28 C62 15730 2026-03-06 200069102 0 068.82 553.28 C62 15730 200069102 2026-03-06 00:42:06 2026-03-06 ARCHIVO