3rd EAGE Workshop on Fiber Optic Sensing for Energy Applications

Not Provided

Not Provided

Not Provided

The distributed acoustic sensing (DAS) technology has shown tremendous potential for high-resolution seismic surveys, owing to its high-density and cost-effective characteristic. However, DAS data often suffer from an exceptionally low signal-to-noise ratio (SNR) due to the presence of various potent noise sources, such as random noise, strong-amplitude erratic noise, and vertical and horizontal stripe noise. To address this challenge, researchers have explored several technologies, for example filtering and deep learning (DL), to enhance the SNR of DAS data. In this study, we present a DL-based denoising framework specifically tailored to attenuate complex DAS noise. Leveraging synthetic DAS data and real noise extracted from field DAS data, we construct a carefully curated dataset for training the designed neural network in a supervised manner, and subsequently, the trained network is served for denoising field DAS data. The experiments demonstrate that the trained network effectively suppresses noise in field DAS data, resulting in a remarkable improvement for the SNR, and the denoising process significantly enhances the detectability of DAS signals.

Not Provided

In this paper, we address the limitations of the popular iterative vertical seismic profiling (VSP) wavefield separation method, which may not achieve high-precision results due to factors like the first break, time-variant wavelet, and strata dip angle. To overcome these challenges, we propose a novel two-stage multi-task U-Network VSP wavefield separation method, leveraging the iterative VSP wavefield separation method, U-Network, and multi-task deep learning. The two-stage multi-task U- Network can effectively extract high-precision downgoing, upgoing, and residual wavefields. Additionally, we introduce synthetic VSP training data automatic generation to facilitate the creation of diverse training datasets. Experimental results using synthetic and actual VSP data demonstrate the efficacy and potential of our proposed method for achieving high-precision VSP wavefield separation.

Hydraulic fracturing is a widely used completion technique for unconventional reservoirs, and monitoring the growth of the fracture network is crucial to ensure safe operations and enhance oil or gas production. The microseismic signals generated during stimulation of induced and natural fractures are monitored for localization purposes. At field-scale, Distributed Acoustic Sensing (DAS) is becoming popular for detecting microseismic events, as it offers a denser spatial sampling. Conducting laboratory hydraulic fracturing experiments with DAS monitoring system provides the opportunity to control variables that cannot be directly measured in the field, allowing to optimize DAS systems and improve their reliability in field monitoring settings. In addition, other governing factors such as confining stress, well geometry, and volume and rate of injected stimulants provides a useful analogy to field completion. In this study, we utilize DAS to monitor hydraulically induced microseismic events in the laboratory within a cubic rock block of 50 cm3in size. A self-reacting triaxial loading frame provides three different confining pressures to generate a true triaxial stress state. Using ISCO pumps we inject the fracturing fluid into a borehole to simulate hydraulic stimulation. DAS fibers are distributed in three directions over six surfaces of the rock block allowing for spatial localization of the microseismic signals. Open corners are designed at each edge of the rock block so that the fiber can form a corner loop to reduce bending losses. With the help of CT-imaging technology we visualize stimulated fractures inside the rock and calibrate microseismic localization results inversed from DAS monitoring. Preliminary test results indicate that the adopted design is effective and reliable for DAS in detecting fracturing signals. Compared to conventional DAS systems, the sampling frequency in this study is increased by about ten times so that high frequency acoustic signals can be better recorded.

In vertical seismic profile (VSP) exploration, distributed optical fiber acoustic sensing (DAS) technology uses optical fiber to simultaneously complete formation strain detection and optical signal transmission. Compared with traditional electronic detectors, this technology has the advantages of higher spatiotemporal accuracy, smaller sampling interval and lower layout cost. However, the borehole seismic recordings explored by DAS are usually contaminated by multi-types of noise with strong energy, such as ringing noise, fading noise, background noise, and optical noise. These noises of different characteristics are caused by the instrument systems or the underground environment, and seriously interfere with the signal continuity of the seismic waves.

Among these noises, the optical noise generated during acquisition is usually manifested as a large-area, strong-energy black and white plaques in the recording, which will completely cover the overlapping seismic reflected signals, causing problems for signal recovery. Obtaining a prior knowledge is the first step in suppressing noise, but at present, not much has been revealed about the production mechanism and characteristics of the optical noise.

Therefore, we try to establish a priori about this noise. Specifically, we study the mathematical modeling method (multiplicative or additive) of this noise based on its statistical distribution, and take the different-trace optical noise sampled from the real borehole recording as examples to draw the conclusion. In order to prove the conclusions obtained, we further use secondary spectrum analysis (complex cepstrum), which is suitable for multiplicative noise analysis, to reflect the frequency differences between seismic waves and optical noise for filtering. According to the frequency standards established by the complex cepstrum, we filter the seismic data interfered by the optical noise with the frequency division point. The filtering results effectively prove the conclusion of the mathematical model analysis we obtained.

Not Provided