This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 195516, “Field Applications of Constrained Multiwell Deconvolution,” by V. Jaffrezic, SPE, Total, K. Razminia, SPE, Imperial College, J. Cumming, Durham University, and A.C. Gringarten, SPE, Imperial College, prepared for the 2019 SPE Europec featured at the 81st EAGE Annual Conference and Exhibition, London, 3–6 June. The paper has not been peer reviewed. This paper applies a new constrained multiwell deconvolution algorithm to two field cases: a gas reservoir with two producers and an oil reservoir with three producers and one injector. Responses given by the constrained multiwell deconvolution are compared with simulations from history-matched reservoir models. By extraction of well and interwell reservoir signatures, multiwell deconvolution, the theoretical background of which is discussed in detail in the complete paper, allows identification of compartmentalization or unanticipated heterogeneities very early in field life, making adjustments possible for field-development plans and future well locations. Introduction Permanent downhole pressure gauges are installed routinely in most new wells. In theory, they provide invaluable data for monitoring well productivity and formation pressure in real time and for calibrating the reservoir-simulation model. In practice, however, the resulting large data sets are usually underexploited, mainly because extraction of meaningful information from interfering wells with conventional well-test-analysis techniques is nearly impossible. Field Case 1 A sandstone dry gas reservoir has been developed with two horizontal wells (Wells 9 and 10) with lengths of approximately 400 m that are 1.6 km apart. The production data set spans 19,800 hours (approximately 2 years and 3 months). No datum correction was applied because the vertical distance between the pressure gauges in the two wells was only 10 m and the correction would be negligible. Data Selection. Because the field case involves dry gas, pressure was converted into single-phase normalized pseudopressure using pressure/volume/temperature data. Depletion is moderate, so use of material balance pseudotime to account for changes in gas properties was unnecessary. Pressure data to be used for multiwell deconvolution had to be selected with care. Flowing pressures are usually too noisy, so buildup data are preferred. Linearity is enforced by only considering buildups with similar pressure derivatives. If these differ at early times, because of changing wellbore and skin effects, a buildup of reference is selected, with elimination of early-time data in all other buildups. Derivatives of the selected buildups, however, may also differ at late times because of interference between wells. These late-time data must be kept because they contain information on interwell connectivity and are necessary to recover constant rate interference responses from multi-well deconvolution. Multiwell Algorithm Setup. The following parameters were selected for the deconvolution process: The first nodes of the well main responses were set at 0.01 hours. The derivative slopes before that time were considered a regression parameter with no constraint applied. The same procedure was used for the interference derivatives except that the first nodes were set at 0.1 hours. All deconvolved derivatives were constrained to converge to a unit slope straight line at late times because the two wells are in communication and the reservoir is closed. The deconvolved derivatives were extended by 1.5 log cycles beyond the test duration, and the constraint of convergence was applied from 22,000 hours. The interference derivatives were constrained to remain below the deconvolved derivatives of the individual wells. The initial pressure was considered a regression parameter with no constraint applied. As a result, the algorithm returned different initial pressure estimates for each well. Default weight parameters were used except for the curvature, where a weight of 0.001 was applied to the main responses and 0.005 to the interferences. These values were selected by trial and error to limit the contribution of the curvature constraint.
Read full abstract