Field Management by Proactive Optimisation of Intelligent Wells - A Practical Approach

Abstract

Abstract Intelligent wells (I-wells) are one of the core elements of the "digital oil field". Numerous publications have described their ability to simultaneously increase recovery and "Add Value". Maximum "Added-Value" is achieved via operation of the wells’ zonal flow-control valves in a manner that respects the field’s optimisation objectives at both a Well (short-term, Reactive control) and a Reservoir (long-term, Proactive control) level. Relatively simple, well-proven algorithms are available for Reactive (mainly well out-flow) control. By contrast, techniques for Proactive control (or reservoir fluid front displacement management) are much less mature; despite them being capable of delivering greater "Added-Value" in many reservoir scenarios. Coupling Genetic Algorithms (GA) with a commercial reservoir simulator is the most commonly used Proactive optimisation technique (Alghareeb et al., 2009, Almeida et al., 2010, Pinto et al., 2012). However, GA has not achieved universal acceptance since, like many other optimisation algorithms, it either fails to find the maximum production improvement and/or requires excessive computational time and resources due to the large number of control variables associated with the optimisation process. This paper discusses the use of the parallel Simultaneous Perturbation Stochastic Approximation (SPSA) algorithm for Proactive optimisation of an I-well in a simplified, real field. The algorithm’s parallel processing capability speeds-up the optimisation process when appropriately employed with a commercial reservoir simulator. A properly tuned SPSA algorithm was found to rapidly converge to a near optimum solution. Algorithm tuning by the engineer was shown to be aided by use of a novel, multi-dimensional, projection scheme to visualise the algorithm’s performance and monitor its progress towards the global optimum solution. Computationally, SPSA is significantly faster (3 – 5 times in the examples tested to-date) than GA. This decrease in processing time can significantly increase the practicality of introducing proactive optimisation for large real field models on a routine basis where it is used alongside the more commonly applied reactive control strategy for managing the individual well performance (Yeten et al., 2002, Grebenkin and Davies, 2012).