Production Surveillance And Optimization With Data Driven Models

Abstract

In conventional practice, individual well oil, gas and water production is only measured on a weekly or monthly basis using shared well test facilities. Oil and gas production from a cluster of wells is hence difficult to manage, leading to late diagnosis of production problems and slow and conservative handling of production constraints. FieldWare Production Universe (FW PU) is a software application developed by Shell International Exploration & Production and Shell Global Solutions International that provides continuous real time estimates of well-by-well oil, water and gas production. FieldWare PU estimates are based on data driven models constructed and updated from production well tests and real time production data. This paper will discuss two extensions of FieldWare PU data driven techniques. The first extension is to apply the data driven models for production optimization. The second extension is the case where no shared test facility for well-by-well production testing is available, and wells can only be tracked by monitoring changes in commingled production flows. For the optimization functionality, the FieldWare PU data driven well models allow the prediction of the changes to overall and individual well production as a result of changes to individual well production chokes, lift-gas rates or other similar set-points. Well setpoints are then computed for optimizing oil and gas production subject to various well and overall production constraints. Data driven techniques for well characterization using commingled production data will be illustrated in three particular production estimation scenarios:individual well production with no shared well testing facility,production from multiple subsea wells sharing a single tie-back pipeline, andproduction from individual subsurface zones of a multi-zonal extended reach Smart Well. FieldWare Production Universe FieldWare Production Universe (FW PU) is a data driven modelling application developed by Shell to address fundamental gaps in the management and surveillance of oil and gas production operations. The development background and early operational experience of FW PU within the Shell Group are described in Poulisse et al. [1] and Cramer et al. [2]. Using data driven models, FW PU essentially provides a "virtual" three phase meter for each well. Earlier references to the potential use of virtual meters includes, for example, van der Geest [4], which is based on physical models. A brief reference to using data driven models for virtual metering, albeit using a less structured neural network approach, is given in Oberwinkler et al. [3]. A recent paper that touches on the potential for data driven modelling is [22] by Stone. Well three phase oil, water and gas production is conventionally measured via the periodic routing of the well to a shared test separator, the "production well testing" process. The duration of the test is normally 6 – 24 hours or longer; the test frequency can vary but is typically weekly, monthly or even less frequent. The usual result of a well test is a set of spot readings and totalized or averaged numbers such as oil production rate, watercut gas-oil-ratio and tubing head pressure. The production of a well is then assumed to be uniformly at the tested production rates between well tests, other then at various intervals when the well is designated to be "closed-in". Sub-normal production rates, unstable production or increases in gas or water production are typically not detected until the next well test. Historically, there have been a number of approaches using well physical models combined with real time wellhead pressures and temperatures to predict 3-phase flow in real time or near real time. In practice, well physical models were found to be difficult to set-up, calibrate and maintain in an operating environment.

Aplicação De Técnicas De Machine Learning E Data Driven Em Poços Inteligentes De Petróleo

“…Vale ressaltar que, devido a característica específica de cada poço (geologia, profundidade, pressão e etc), não existe uma metodologia única e seria improvável encontrar uma solução de fácil generalização. Goh et al (2007) usou a técnica data driven em duas situações. A primeira delas foi para estimar a previsão em tempo real da produção de petróleo, água e gás de cada poço de um reservatório, dada uma mudança na produção em um dos poços.…”