New approach for interpreting pressure and flow rate data from permanent downhole gauges, least square support vector machine approach

Abstract

Permanent downhole gauges (PDGs) are permanently installed in oil and gas wells to monitor real-time downhole pressure, temperature and sometimes flow rate. Recording these variables continuously plays a key role in reservoir analysis. Conventional welltesting methods as one group of approaches for reservoir characterization are generally used in a way that interpreting the obtained reservoir pressure (reservoir response) will be easy. For this target to be achieved welltesting often deals with constant flow rate sections of data in a short period of time. As an alternative approach, machine learning deals with large volumes of data (which are recorded in a long period of times) to find the relationship among the related variables (Liu and Horne, 2013a). Here, we aim to describe the usage of the least square support vector machine (LS-SVM) as a machine learning method for analyzing data from permanent downhole gauges. This approach is a tool for analysis of data that are in the presence of noise, gaps and outliers which reveals another advantage of using LS-SVM instead of conventional well-testing methods in which designed tools in controlled environments are needed. Last but not least, this method is one of the few ways which can use the flow-rate data with uncertainty. In this document, the application of radial basis function (RBF) as a function approximation for mapping the data to a high dimensional feature space is investigated. The LS-SVM method proved to be successful in predicting pressure especially when the wellbore/reservoir model contains a nonlinear pattern or when the pressure derivative curve has a nonlinear trend. Since PDG records data during production and not during an imposed flow rate condition, the existence of nonlinear features is common which proves the applicability of the presented method.