Technique Detects Well Interference Using Long-Term Pressure Data

Abstract

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 206354, “Well-Interference Detection From Long-Term Pressure Data Using Machine Learning and Multiresolution Analysis,” by Dante Orta Aleman, SPE, and Roland Horne, SPE, Stanford University. The paper has not been peer reviewed. Methods such as interference tests and tracers have been developed to obtain knowledge of reservoir heterogeneity and connectivity from dynamic data. However, detecting well connectivity using interference tests requires long periods with stable reservoir pressure and constant flow-rate conditions. Conversely, the long duration and high frequency of well-production data have high value for detecting connectivity if noise, abrupt changes in flow rate, and missing data are considered. In this work, a methodology to detect interference from long-term pressure and flow-rate data was developed using multiresolution analysis in combination with machine-learning algorithms. Introduction Multiple methodologies have been proposed to detect well connectivity using production data. Most such methods apply a statistical or physical model to relate production and injection rates. In previous works, pressure data rarely has been used and data were fitted to a predefined mathematical model with strong physical assumptions. The approach of the current work is to infer connectivity by estimating the pressure response of a well to a second “interfering” well. Because the pressure response at a given well is the convolution of all the connected wells’ flow rates with the reservoir response, the methodology attempts to deconvolve the flow rates and assign a pressure response to each. Previous work showed that, by using wavelet decomposition and deep learning, capture of the pressure response of a single well using only flow-rate data as input is possible. The complete paper expands on the idea of using the multiresolution properties of wavelets to simplify the flow-rate/pressure response of more than one well and to capture that response by using a machine-learning model with a flexible function representation.