Probabilistic Well-Time Estimation Revisited

Abstract

Summary Probabilistic estimation of well duration has been common practice for more than a decade; many papers have been written on the subject, and commercial software is available for the purpose. Is the subject, therefore, mature? The authors suggest that this is not the case and show that several essential aspects of both data characterization and probabilistic analysis have been overlooked in previous contributions. In 2007, Nexen began a study with the objective of improving our process for well-time estimation. Its workscope was to Assemble a statistically significant historical well database.Develop a consistent definition of nonproductive time (NPT) as it relates to the original approved-for-expenditure (AFE) timings.Reanalyze the historical well database for NPT from the original daily drilling reports (DDRs), on the basis of the definition above.Decide how NPT is best characterized (that is, determine the correct choice of input variables).Determine the associated occurrence frequencies and probability-density-function (PDF) parameters from the historical database.Validate the probabilistic model, by comparison of program predictions against the historical data set.Work with a software provider to implement any necessary changes in a commercially available product. A database of 118 central North Sea wells was reanalyzed independently for NPT from the original DDRs. These considerably underestimated the true NPT, by 19.7% on average. Train wrecks (mechanical NPT events of more than 2.5 days) were only 4% by number but contributed 50% of NPT by duration. It was found that "ordinary" mechanical NPT, train wrecks, waiting on weather (WOW) in open water, and WOW with riser connected are all statistically distinct, with very different occurrence frequencies and PDFs. Earlier workers did not observe this distinction or properly validate their models against the well database. Therefore, it is impossible to obtain historically accurate probabilistic well-time predictions (i.e., that are consistent with the historical database) using the previous state of the art. Moreover, such predictions will generally be underestimates. This paper describes an accurate method that overcomes previous limitations. While central North Sea data are used, the analysis techniques are not area specific, and the method may be applied easily to other areas in the oil field.