Robust Regression and Band Switching to Improve DAS Flow Estimates

Abstract

Within the Oil and Gas industry the use of Acoustics data for flow rate estimation is increasingly being explored. One technique is to consider the total spectral power of the signal within a specific frequency range, known as an FBE. The FBE, along with measured Flow rates, can then be used to build a simple regression model to estimate the flow rate. We collect acoustic data using Distributed Acoustic Sensing, DAS, and find that the recorded FBE generally contains some corrupted data and outliers. This may be due to well shut-in periods or other physical phenomena, or it may be due to issues in the DAS recording itself. These outliers can have a detrimental effect on the calibration of any predictive model and lead to biased flow predictions. We combat this by calibrating out model using Robust Regression techniques, such as Least Absolute Deviation, which are less influenced by outliers. Another practical concern is choosing the correct frequency band for the FBE. This can be done by evaluating the model performance on a training set, however we find that the signal quality within a band can diminish over time necessitating a change in the band used. Our challenge is to find a way to identify when a band is likely to be giving poor predictions. We do this by looking at the ratios between different FBE bands, we find that under normal conditions these are highly correlated, however for certain bands this correlation is lost over time. This can be used to determine when it is time to switch to use a different band. This paper contains the motivation and results of these techniques as they are applied to flow prediction in a gas producing well which has been part of a long-term flow monitoring project.