With current trends towards intelligent wells and fields, continuous bottomhole well pressure monitoring is becoming the norm in new field developments. Theoretically, this should allow operators to better control well performance and address problems before they become irreversible. In practice, the need to interpret the raw information provided by permanent gauges, and the lack of manpower or expertise for doing so prevents real time intervention. Therefore, some sort of automatic interpretation and alarm system is required to benefit from the full potential of downhole permanent gauges.

A new deconvolution algorithm developed at Imperial College by von Schroeter, Hollaender and Gringarten makes the development of such tools possible. Contrary to deconvolution algorithms previously published in the literature, the Imperial College algorithm provides stable results. Different implementations of the algorithm have been reported in the literature and as a result, deconvolution is rapidly becoming a significant step in well test analysis. Benefits are numerous: it provides information over a radius of investigation that corresponds to the total duration of the production period, which can be orders of magnitude greater than that available from individual flow periods used for conventional well test analysis; it can reduce considerably the required duration of an extended well test, by revealing boundaries before they become visible in individual flow periods; it allows differentiating between alternative interpretation models; and it may be the only way to interpret well test data, for instance in the case of multilateral horizontal wells in low permeability reservoirs.

The current deconvolution algorithm is only valid for a single well. In order to exploit fully the potential of deconvolution, the algorithm must be able to handle multiwell systems, where pressure interference occurs, and provide a deconvolved derivative for every interfering well. This has been the subject of the 2005-2007 research project at Imperial College sponsored by a consortium of 14 companies*.  It was found that, contrary to early expectations, a direct extension of the single well deconvolution algorithm was impractical beyong two wells, because of the number of control parameters required, and that a more promising approach was to reformulate the deconvolution problem within the framework of Bayesian statistics. The 2005-2007 research work concentrated on analytic Bayesian solutions which unfortunately require simplifying assumptions which are difficult to check. A better approach is now proposed, which is to use Markov chain Monte Carlo (MCMC) methods.

The primary objective of the research is to complete the deconvolution work started in the 2005-2007 Imperial research project, using Bayesian statistics in order to obtain a robust multiwell deconvolution algorithm that can be used by practicing engineers in a wide variety of field conditions to control and optimise well performance.

A second objective is to investigate the use of Bayesian statistics for well test interpretation model identification, in order to obtain estimates of reservoir properties and future flow/pressure behaviours with stated levels of accuracy.

*Saudi Aramco, BG, BHP, BP, Chevron , ConocoPhillips, ENI, Gaz de France, Occidental Petroleum, Petro SA, Schlumberger, Shell, Total, Weatherford