Abstract

Distributed fiber-optic sensing, including distributed acoustic sensing (DAS) and distributed temperature sensing (DTS), has gained interest in a range of applications in oil & gas fields. DTS and DAS use optical time domain reflectometry (OTDR) measurement techniques in which an incident light pulse is coupled into an optical fiber and backscattered light sampled. Continuous analyses of the backscattered signal from successive incident pulses can provide dynamic profiles of both temperature and acoustics (dynamic strain) continuously over time and distance along the fiber.In a production environment, DAS and DTS measurement can provide wellbore characterization, flow profiling and allocation over short duration or through permanently deployed systems, with information over different frequencies sensitive to gas/liquid flow rates, flow regime, fluid transport and interfaces, flow through ICVs and GLVs, et al. The volume of combined DAS and DTS data can be quite significant over a short period of investigation time per well, and overwhelmingly large for a field with a number of wells. Prediction and characterization based on these data using traditional techniques are often qualitative and becomes inadequate. We have developed a deep learning based framework to predict flow profiling and allocation using combined DAS and DTS data. Specifically we train a convolutional neural network mode using both the DAS and DTS data collected from several injectors and producing wells, to predict flow regime labels such as the baseline production flow, flow states associated with Injection and slug, shut in, et al. We investigate the frequency sensitivity of DAS measurement in predicting the flow regimes. Field results obtained from applying the trained deep learning network to new test data indicate the method is robust and accurate.

This content is only available via PDF.
You can access this article if you purchase or spend a download.