Use of Radial Basis Function Networks for Efficient Well Production Allocation

Abstract

Well production allocation is the cornerstone of reservoir surveillance and sound reservoir management. The apparent simplicity of the allocation process often results in an underestimation of its critical importance. However, the accuracy of the production rates allocation has an overwhelming impact on the company's ability to use sound data and perform model-driven analytics. As a result, the reliability of production forecasts, reserves estimates, and production system optimization efforts are affected by the selected allocation approach. A common approach to well production allocation is based on the use of well tests closest in time to the point of interest. It assumes stable operating conditions and gradual changes in fractions of produced fluids. These assumptions rarely reflect reality and therefore lead to large allocation errors. Use of more sophisticated solutions, such as data-driven and model-driven integrated well-reservoir tools pose different challenges due to the constant need for time-consuming updates. In this paper, we present a quick and efficient approach for production data allocation based on single layer Radial Basis Function Network - a variation of Artificial Neural Network. The procedure takes advantage of full well test dataset and can be effectively used in real time. We show that this approach does not suffer from the limitations of the more common approaches while delivering improved results.