Non-parametric regression and neural-network infill drilling recovery models for carbonate reservoirs

Citation
Ch. Wu et al., Non-parametric regression and neural-network infill drilling recovery models for carbonate reservoirs, COMPUT GEOS, 26(8), 2000, pp. 975-987
Citations number
10
Categorie Soggetti
Earth Sciences
Journal title
COMPUTERS & GEOSCIENCES
ISSN journal
00983004 → ACNP
Volume
26
Issue
8
Year of publication
2000
Pages
975 - 987
Database
ISI
SICI code
0098-3004(200010)26:8<975:NRANID>2.0.ZU;2-H
Abstract
This work introduces non-parametric regression and neural network models fo r forecasting the infill drilling ultimate oil recovery from reservoirs in San Andres and Clearfork carbonate formations in West Texas. Development of the oil recovery forecast models helps understand the relative importance of dominant reservoir characteristics and operations variables, reproduce r ecoveries for units included in the database, forecast recoveries for possi ble new units in similar geological settings, and make operations decisions . The variety of applications demands the creation of multiple recovery for ecast models. One of the significant constraints for the model development is the limited number of field data that are inexact and often exhibit unce rtain relationships. The inexact and uncertain relationship may also encomp ass a large number of possible independent variables. This situation mandat es proper selection of independent variables for the infill drilling recove ry model. Non-parametric regression and multivariate principal component an alysis are used to identify the dominant and the optimum number of independ ent variables. The advantage of the non-parametric regression is easy to us e and can quickly provide results that reveal the dominant independent vari ables and relative characteristics of the relationships. The disadvantage i s retaining a large variance of forecast results for a particular data set. The insight of interdependency of the variables gained in non-parametric r egression and multivariate principal component analysis is employed to deve lop an effective neural network. The neural network infill drilling recover y model is capable of forecasting the oil recovery with less error variance . This work shows that a multiple use of various modeling techniques may pr ovide a healthy interaction between the different approaches and thereby, a better oil recovery forecast. (C) 2000 Elsevier Science Ltd. All rights re served.