Trezzi, Stefano (Schlumberger) | Glushchenko, Anna (Schlumberger) | Vasilyev, Pavel (Schlumberger) | Al Bulushi, Abdullah Mubarak Khamis (Schlumberger) | Cox, Edward (SNOC) | Al Hamadi, Masoud Ahmed (SNOC) | Stewart, Neil (SNOC)
This study summarizes the efforts taken to provide reliable structural delineations offshore the northern United Arab Emirates, an area where previous attempts failed to provide purposeful results.
With the employment of latest acquisition and processing techniques, a new high-density full azimuth volume shows clear uplift over legacy results. Structures never detected before are imaged, contributing to the de-risking of future well placement. These results are indicative of the challenges when acquiring and processing seismic data in the Northern Emirates.
The use of latest technologies was required to overcome several geophysical challenges, such as complex near surface, near-vertical thrust sequences and multiple faulting planes. All these elements contributed in generating extreme noise contamination, where a fast-varying geology with high dips made separation of primaries from noise one of the most difficult tasks.
Key pre-imaging technologies such as near surface characterization, 5D regularization to radial symmetry, as well as a tomographic velocity model building approach with iterative inputs from interpreters have been fundamental to converge to a solid velocity model that allows for a reliable structural imaging.
The work is particularly relevant for offshore exploration targets in the Northern Emirates, a region that has recently seen a growth in interest from local and international operators.
A state-of-the-art, high-density full azimuth seismic dataset was acquired over a producing field southwest Abu Dhabi to determine optimum high-resolution acquisition and processing parameters. The existing 3D seismic data was acquired in 1994-95 with low fold coverage. The central part of the new seismic volume has good well control to develop, optimize, and QC processing workflows. The new dataset utilised the system with single-sensor geophone accelerometers, point source DX80 vibrators using the Maximum Displacement sweep design with an enhancement of low frequencies as such a broadened amplitude spectrum. The key specifications for seismic acquisition include full azimuth high-density single-source / single-sensor data, long offsets and a broadband vibroseis sweep that retains low frequencies down to 3Hz. The single-sensor single-source technology and the full azimuth dense geometry (5x5m grid) was designed for an optimum subsurface imaging for this survey. The volume was processed through anisotropic prestack time and depth migration followed by poststack wavelet processing and spectral balancing.
Decline curve analysis is widely used in industry to perform production forecasting and to estimate reserves volumes. A useful technique in verifying the validity of a decline model is to estimate the Arps decline parameters, the loss ratio and the b-factor, with respect to time. This is used to check the model fit and to determine the flow regimes under which the reservoir produces. Existing methods to estimate the b-factor are heavily impacted by noise in production data. In this work, we introduce a new method to estimate the Arps decline parameters.
We treat the loss ratio and the b-factor over time as parameters to be estimated in a Bayesian framework. We include prior information on the parameters in the model. This serves to regularize the solution and prevent noise in the data from being amplified. We then fit the parameters to the model using Markov chain Monte Carlo methods to obtain probability distributions of the parameters. These distributions characterize the uncertainty in the parameters being estimated. We then compare our method with existing methods using simulated and field data.
We show that our method produces smooth loss ratio and b-factor estimates over time. Estimates using the three-point derivative method are not matched with data, and results in biased estimates of the Arps parameters. This can lead to misleading fits in decline curve analysis and unreliable estimates of reserves. We show that our technique helps in identification of end of linear flow and start of boundary dominated flow. We use our method on simulated data, with and without noise. Finally, we demonstrate the validity of our method on field cases.
Fitting a decline curve using the loss ratio and b-factor plots is a powerful technique that can highlight important features in the data and the possible points of failure of a model. Calculating these plots using the Bourdet three-point derivative induces bias and magnifies noise. Our analysis ensures that this estimation is robust and repeatable by adding prior information on the parameters to the model and by calibrating the estimates to the data.
Li, Baoyan (Baker Hughes, a GE Company) | Arro, Roberto (Baker Hughes, a GE Company) | Thern, Holger (Baker Hughes, a GE Company) | Kesserwan, Hasan (Baker Hughes, a GE Company) | Jin, Guodong (Baker Hughes, a GE Company)
For NMR logging of hydrocarbon bearing formations, the inversion of T2 echo trains is a critical pre-processing step to compute porosity, permeability, and fluid saturations of formations. To accurately and efficiently invert NMR measurement data, a new inversion method is presented to compute the optimal solution of T2 distribution with a unique optimal regularization factor for NMR logging data processing. This inversion method has no assumption about the white noise of a T2 echo train.
A new supplementary nonlinear equality constraint was introduced to optimize the solution of T2 distribution by explicitly taking into account the measured noise of a measured T2 echo train. An efficient iterative algorithm has been developed to solve the nonlinear optimization problem defined in the new inversion method. An initial-guess solution of the regularization factor was proposed for accelerating the searching process of the regularization factor.
The new inversion method has been verified with synthetic T2 echo train data and applied to process T2 echo train data of core samples of carbonate and Berea sandstone formations that are saturated with different fluids. This method has also been compared with conventional methods. The testing and comparison results show that: The optimal solution of T2 distribution from the new inversion method has a unique solution that is independent of the pre-selected values of regularization factor, so does the regularization factor. The optimal solutions T2 distribution and regularization factor will be convergent to their true solutions when the SNR of echo train data becomes sufficiently high. The computation cost for searching the optimal solutions of T2 distribution and regularization factor using the new nonlinear optimization algorithm is only a few iterations. The initial-guess solution of the regularization factor is more close to the solution determined from the S-curve method, which could be higher than the optimal solution of the regularization factor searched in the new inversion method.
The optimal solution of T2 distribution from the new inversion method has a unique solution that is independent of the pre-selected values of regularization factor, so does the regularization factor. The optimal solutions T2 distribution and regularization factor will be convergent to their true solutions when the SNR of echo train data becomes sufficiently high.
The computation cost for searching the optimal solutions of T2 distribution and regularization factor using the new nonlinear optimization algorithm is only a few iterations.
The initial-guess solution of the regularization factor is more close to the solution determined from the S-curve method, which could be higher than the optimal solution of the regularization factor searched in the new inversion method.
Quantification of oil and water saturations in unconventional reservoir rock using traditional single-scan NMR methods like CPMG is difficult because hydrocarbon and water in small pores both relax at very short times and are often indiscernible. Recent developments in the simultaneous acquisition and inversion of spin-lattice (T1) and spin-spin (T2) relaxation creates an intensity map of T1 versus T2 relaxation time distributions. Although the T1 vs T2 “map” provides an easy-to-understand visual tool for interpreting saturations and fluid-rock interactions in shale, it lacks some of the robustness to make comparisons with direct measurement techniques like Dean-Stark or retort.
This paper proposes an alternative workflow and interpretation strategy applicable for low-field (2-23MHz) instruments that measures T1 on an “as received” sample and T2 on the same samples in the “as-received” and “fully saturated” state. A key step in this workflow is controlling the inversion parameters that generate the T1 and T2 distributions in order to extract more information from the NMR measurements.
Comparison of water volume measurements on standard 3.7 cm diameter core plugs shows the method developed using stand-alone T1 and T2 measurements agrees with Dean-Stark within 0.3mL almost 60% of the time. Sometimes signal reprocessing results in more peaks than can be logically assigned and these instances provide merit for spot-checking the interpretation with direct measurements. The ratio of T1 to T2 relaxation times is useful as a proxy for hydrocarbon mobility and relates to the amount of organic matter in the sample.
Nuclear Magnetic Resonance (NMR) measurements are a valuable tool for quantifying total porosity and saturations in conventional reservoirs. Oilfield NMR measurements rely on measuring the resonant properties of precessing hydrogen nuclei, specifically 1H. Measuring the 1H resonant properties is called proton NMR and it's useful because protons are ubiquitous in reservoir fluids: oil, hydrocarbon gas, and brine. The most common NMR measurements for porosity and saturation quantification rely on spin-lattice (T1), or spin-spin (T2) relaxation with the NMR T2 measurement being the most commonly used for laboratory and downhole measurements.
The velocity model is of a great importance for geological as well as structural properties of complex structure such as gas cloud. Instead of ray-based techniques, eikonal wavefield tomography can provide a higher resolution velocity model for seismic images. We have implemented first break travel time tomography to enhance the initial velocity model for seismic full waveform inversion (FWI) for better imaging rather than guess initial velocity model for FWI. The First-break travel time concept is based on the eikonal equation, relies on inversion to resolve the complex gas cloud imaging. It allows not only the receivers but the shots to change position along the ray path. Tomography results are useful particularly significant in the presence of noise, scattering in the data. We have implemented this approach on marmousi as well as gas cloud model and output are used as input velocity model for FWI and results of proposed approach is more robust than the traditional with faster convergence.
I examine the basis of slow convergence of tomographic full waveform inversion (TFWI) and discover that the reason behind it is the unbalanced effects of amplitudes and phase in the design of the regularization term. This imbalance results in a strong reliance of the kinematic updates on the amplitude fitting, slowing down the convergence. To mitigate the problem I propose two modifications to the tomographic inversion. First, by modifying the regularization term to focus more on the phase information, and second, simultaneously updating the source function for modeling. The adjustments reduce the gradient artifacts and allow for explicit control over the amplitudes and phases of the residuals.
Tomographic full waveform inversion (
The modeling operator is able to match the observed data by extending the velocity model with the proper axis, no matter what the accuracy of the initial model is, by using kinematic information from the extended axis with disregard to the occurrence of cycle skipping. The inversion is set up to extract all the essential information from the virtual axes and smoothly fold them back into their original, nonextended form of the model. The kinematic and dynamic information of the data were successfully inverted with exceptional robustness and precision.
Even though cycle-skipping is not an issue with TFWI, this method creates its own challenges, which are; its high computational cost and the big number of iterations that it needs (
Two adjustments to TFWI are proposed to reduce the slow convergence and allow for more control of the ratio between amplitude and phase. These adjustments are consistent in the framework of TFWI and allow for an accurate calculation of the gradient in the data space. The adjustments were tested and resulted in a reduction in the kinematic artifacts in the gradient.
Shin, Seungwook (Center for Gravity, Electrical, and Magnetic Studies, Colorado School of Mines, Golden, Colorado) | Li, Yaoguo (Center for Gravity, Electrical, and Magnetic Studies, Colorado School of Mines, Golden, Colorado) | Park, Samgyu (Korea Institute Geoscience and Mineral Resource (KIGAM), Daejeon, South Korea) | Cho, Seong-Jun (Korea Institute Geoscience and Mineral Resource (KIGAM), Daejeon, South Korea)
Summary It has long been that we may be able to differentiate between different lithology units from the dependence of induced polarization (IP) signal on the mineral composition and pore space distributions. The commonly used, knowledge-driven approach based on equivalent circuit analyses, however, tends to be challenged with severe ambiguities. To overcome this difficulty, we experiment with supervised machine learning based on data-driven approach. We present a study on the feasibility of using artificial neural networks to predict lithology from spectral IP data. We show that reliable results can be obtained when the algorithm is trained to obtain a stable set of weights by applying regularization or by terminating the training based on a cross-validation criterion.
Cheng, Liang (China University of Petroleum-Beijing) | Wang, Shangxu (China University of Petroleum-Beijing) | Yuan, Sanyi (China University of Petroleum-Beijing) | Wang, Guanchao (China University of Petroleum-Beijing) | Yu, Zizhao (China University of Petroleum-Beijing) | Deng, Li (China University of Petroleum-Beijing)
In this study, we investigate a mixed norm spatiotemporal regularization that combines L2-norm and Lp-norm on the range 0<p<1 as a constraint to acquire a sparser and more stable deconvolution solution among multi-trace reflectivity or impedance. The spatiotemporal regularization is described as an L2 regularization along the structure direction and an Lp regularization along the time direction in the mixed norm deconvolution. Therefore, it can act as a band-pass filter along the structure or time directions and enhance the continuity among different traces in these spatiotemporal directions. As well, Lp-norm can suppress the high-wavenumber components and majorly restore the inherent sparsity of the estimated reflectivity or impedance. Moreover, this mixed norm spatiotemporal regularization can help stabilizing deconvolution and accurately exploring the spatiotemporal correlation among traces. On basis of the objective function, the inverted equation constrained by mixed norm is derived implemented by minimizing a data misfit term, an Lp-norm term along the time direction and an L2-norm term along the structural direction. In view of the complexity of the inverse equation, we use repeated weighted iterative (RWI) method to solve the problem. Furthermore, a simple model with one dip bed clearly illustrates that the inverse method can help to contribute to a high resolution and signal-to-noise ratio inverse result.
Presentation Date: Wednesday, October 17, 2018
Start Time: 8:30:00 AM
Location: 206A (Anaheim Convention Center)
Presentation Type: Oral
Yang, Yang (School of Electronic and Information Engineering, Xi’an Jiaotong University, China) | Gao, Jinghuai (School of Electronic and Information Engineering, Xi’an Jiaotong University, China) | Zhang, Guowei (School of Electronic and Information Engineering, Xi’an Jiaotong University, China) | Zhu, Xiangxiang (School of Mathematic and Statistics, Xi’an Jiaotong University, China)
Seismic random noise always exists in seismic data acquisition. Dictionary learning methods are the effective tools to find a typical sparse representation. It can be introduced to reduce the seismic random noise. According to the geometric structures of 2D seismic data, the effective seismic data are also low-rank in time-space domain. Therefore, the effective seismic data can be models as a linear combination of a few elements from a learned dictionary. In this paper, we propose a new dictionary learning method to reduce the 2D seismic random noise in time-space domain by incorporating self-pace (SP) learning method with the nonnegative dictionary learning. This method is implemented by sequentially including matrix elements into nonnegative matrix factorization (NMF) training from easy to complex. It can void bad local minima and obtain a high SNR seismic data. The effectiveness of the proposed self-pace nonnegative matrix factorization (SPNMF) method is demonstrated by the synthetic data and the field data.
Presentation Date: Tuesday, October 16, 2018
Start Time: 9:20:00 AM
Location: Poster Station 9
Presentation Type: Poster