The size of the individual seismic surveys has increased over the last decade, along with the generation of megamerge and even larger, what some operators call “gigamerge” surveys. The number of useful attribute volumes has also increased, such that interpreters may need to integrate terabytes of data. During the past several years, various machine learning methods including unsupervised, supervised and deep learning have been developed to better cope with such large amounts of information. In this study we apply several unsupervised machine learning methods to a seismic data volume from the Barents Sea, on which we had previously interpreted shallow high-amplitude anomalies using traditional interactive interpretation workflows. Specifically, we apply k-means, principal component analysis, self-organizing mapping and generative topographic mapping to a suite of attributes and compare them to previously generated P-impedance, porosity and Vclay displays, and find that self-organized mapping and the generative topographic mapping provide additional information of interpretation interest.
In the late 1980s, seismic facies analysis was carried out on 2D seismic data by visually examining the seismic waveforms that can be characterized by their amplitude, frequency and phase expression. Such information would be posted on maps and contoured to generate facies maps. As seismic data volumes increased in size with the adoption of 3D seismic data in the early 1990s, interpreters found that 3D seismic attributes highlighted patterns that facilitated the human recognition of geologic features on time and horizon slices, thereby both accelerating and further quantifying the interpretation. More recently, computer-assisted seismic facies classification techniques have evolved. Such methods or workflows examine seismic data or their derived geometric, spectral, or geomechanical attributes and assign each voxel to one of a finite number of classes, each of which is assumed to represent seismic facies. Such seismic facies may or may not represent geologic facies or petrophysical rock types. In this workflow, well log data, completion data, or production data are then used to determine if a given seismic facies is unique and should be lumped (or “clustered”) with other similar facies determined from attributes with similar attribute expression.
The petro-elastic model (PEM) represents an integral component in the closed-loop calibration of integrated four-dimensional (4D) solutions incorporating time-lapse seismic, elastic and petrophysical rock property modeling, and reservoir simulation. Calibration of the reservoir simulation model is needed so that it is not only consistent with production history but also with the contemporaneous subsurface description as characterized by time-lapse seismic. The PEM requires dry rock properties in its description, which are typically derived from mechanical rock tests. In the absence of those mechanical tests, a small data challenge is posed, whereby all necessary data is not available but the value of reconciling seismic attributes to simulated production remains. A seismic inversion-constrained n-dimensional metaheuristic optimization technique is employed directly on three-dimensional (3D) geocellular arrays to determine elastic and density properties for the PEM embedded in the commercial reservoir simulator.
Ill-posed dry elastic and density property models are considered in a field case where the seismic inversion and petrophysical property model constrained by seismic inversion exist. An n-dimensional design optimization technique is implemented to determine the optimal solution of a multidimensional pseudo-objective function comprised of multidimensional design variables. This study investigates the execution of a modified particle swarm optimization (PSO) method combined with an exterior penalty function (EPF) with varied constraints. The proposed technique involves using n-dimensional design optimization to solve the pseudo-objective function comprised of the PSO and EPF given limited availability of constraints. In this work, an examination of heavily and reduced-order penalized metaheuristic optimization processes, where the design variables and optimal solution are derived from 3D arrays, is conducted so that constraint applicability is quantified. While the process is examined specifically for PEM, it can be applied to other data-limited modeling techniques.
Integration of time-lapse seismic data into dynamic reservoir model is an efficient process in calibrating reservoir parameters update. The choice of the metric which will measure the misfit between observed data and simulated model has a considerable effect on the history matching process, and then on the optimal ensemble model acquired. History matching using 4D seismic and production data simultaneously is still a challenge due to the nature of the two different type of data (time-series and maps or volumes based).
Conventionally, the formulation used for the misfit is least square, which is widely used for production data matching. Distance measurement based objective functions designed for 4D image comparison have been explored in recent years and has been proven to be reliable. This study explores history matching process by introducing a merged objective function, between the production and the 4D seismic data. The proposed approach in this paper is to make comparable this two type of data (well and seismic) in a unique objective function, which will be optimised, avoiding by then the question of weights. An adaptive evolutionary optimisation algorithm has been used for the history matching loop. Local and global reservoir parameters are perturbed in this process, which include porosity, permeability, net-to-gross, and fault transmissibility.
This production and seismic history matching has been applied on a UKCS field, it shows that a acceptalbe production data matching is achieved while honouring saturation information obtained from 4D seismic surveys.
Time-lapse seismic monitoring is a powerful technique for reservoir management and the optimization of hydrocarbon recovery. In time-lapse seismic datasets, the difference in seismic properties across different vintages enables the detection of spatio-temporal changes in saturated properties and structure induced by production. The main objectives are (1) to identify bypass pay zones in time-lapse seismic data for the deepwater Amberjack field, located in the Gulf of Mexico, (2) confirm the identified bypass pay zones in the results of reservoir simulation, and (3) recommend well planning strategies to exploit these bypassed resources.
A high-fidelity seismic-to-simulation 4D workflow that incorporates seismic, petrophysics, petrophysical property modeling, and reservoir simulation was employed, which leveraged cross-discipline interaction, interpretation, and integration to extend asset management capabilities. The workflow addresses geology (well log interpretation and framework development), geophysics (seismic interpretation, velocity modeling, and seismic inversion), and petrophysical property modeling (earth models and co-located co-simulation of petrophysical properties with P-impedance from seismic inversion). An embedded petro-elastic model (PEM) in the reservoir simulator is then used to affiliate spatial dry rock properties with saturation properties to compute dynamic elastic properties, which can be related to multi-vintage P-impedance from time-lapse seismic inversion. In the absence of the requisite dry rock properties for the PEM, a small data engine is used to determine these absent properties using metaheuristic optimization techniques. Specifically, two particle swarm optimization (PSO) applications, including an exterior penalty function (EPF), are modified resulting in the development of nested and average methods, respectively. These methods simultaneously calculate the missing rock parameters (dry rock bulk modulus, shear modulus, and density) necessary for dynamic, embedded P-impedance calculation in the history-constrained reservoir simulation results. Afterward, a graphic-enabled method was devised to appropriately classify the threshold to discriminate non-reservoir (including bypassed pay) and reservoir from the P-impedance difference. Its results are compared to unsupervised learning (k-means clustering and hierarchical clustering). From seismic data, one can identify bypassed pay locations, which are confirmed from reservoir simulation after conducting a seismic-driven history match. Finally, infill wells are planned, and then modeled in the reservoir simulator.
Reservoirs and the lateral seal of stratigraphic traps are controlled by the depositional environment or diagenesis. The recognition of facies and lithology from seismic attributes is an effective approach for identifying stratigraphic traps related to the depositional environment. In this paper, the occurrence of stratigraphic traps related to depositional environment in Permian aeolian clastics and Jurassic carbonate-evaporites was studied. To identify these stratigraphic traps, multiple seismic attributes were classified using supervised and unsupervised artificial neural networks (ANNs), which allowed the recognition of seismic facies and lithology.
Neural networks are a powerful classification technique, which incorporates multiple attributes into a number of classes to identify sedimentary facies. Two algorithms comprising supervised and unsupervised neural networks are commonly implemented. With a supervised learning algorithm, prior information such as typical facies at the control wells are required to train the multilayer perceptron (MLP) network. With an unsupervised algorithm, only seismic data is input to the neural network, and competitive-learning techniques are employed to classify or self-organize the data based on its internal characteristics. Without prior information, the output classes are not labeled with lithofacies. According to the availability of prior information, supervised and unsupervised learning were applied to recognize dune-playa and carbonate-evaporite combinations, respectively. To characterize the depositional environments, joint interpretation with a geological model is necessary for both supervised and unsupervised classification.
Two major findings have been derived from this work. First, the learning technology based on ANNs is effective to recognize sedimentary facies. The microfacies and lithologies identified by both supervised and unsupervised ANNs are very consistent with the drilled wells. Second, the recognition of depositional facies and lithology can characterize the stratigraphic traps in the study areas. Lateral seal plays a key role in stratigraphic traps. Playa siltstone and tight lagoonal limestone constitute the lateral seal in dune-playa and carbonate-evaporite combinations, respectively.
Seismic attributes are a well-established method for highlighting subtle features buried in seismic data in order to improve interpretability and suitability for quantitative analysis. Seismic attributes are a critical enabling technology in such areas thin bed analysis, 3D geobody extraction, and seismic geomorphology. When it comes to seismic attributes, we often suffer from an "abundance of riches" as the high dimensionality of seismic attributes may cause great difficulty in accomplishing even simple tasks. Spectral decomposition, for instance, typically produces 10's and sometimes 100's of attributes. However, when it comes to visualization, for instance, we are limited to visualizing three or at most four attributes simultaneously.
My co-authors and I first proposed the use of latent space analysis to reduce the dimensionality of seismic attributes in 2009. At the time, we focused upon the use of non-linear methods such as self-organizing maps (SOM) and generative topological maps (GTM). Since then, many other researchers have significantly expanded the list of unsupervised methods as well as supervised learning. Additionally, latent space methods have been adopted in a number of commercial interpretation and visualization software packages.
In this paper, we introduce a novel deep learning-based approach to latent space analysis. This method is superior in that it is able to remove redundant information and focus upon capturing essential information rather than just focusing upon probability density functions or clusters in a high dimensional space. Furthermore, our method provides a quantitative way to assess the fit of the latent space to the original data.
We apply our method to a seismic data set from the Canterbury Basin, New Zealand. We examine the goodness of fit of our model by comparing the input data to what can be reproduced from the reduced dimensional data. We provide an interpretation based upon our method.
Alkinani, Husam H. (Missouri University of Science and Technology) | Al-Hameedi, Abo Taleb T. (Missouri University of Science and Technology) | Dunn-Norman, Shari (Missouri University of Science and Technology) | Flori, Ralph E. (Missouri University of Science and Technology) | Alsaba, Mortadha T. (Australian College of Kuwait) | Amer, Ahmed S. (Newpark Technology Center/ Newpark Drilling Fluids)
Oil/gas exploration, drilling, production, and reservoir management are challenging these days since most oil and gas conventional sources are already discovered and have been producing for many years. That is why petroleum engineers are trying to use advanced tools such as artificial neural networks (ANNs) to help to make the decision to reduce nonproductive time and cost. A good number of papers about the applications of ANNs in the petroleum literature were reviewed and summarized in tables. The applications were classified into four groups; applications of ANNs in explorations, drilling, production, and reservoir engineering. A good number of applications in the literature of petroleum engineering were tabulated. Also, a formalized methodology to apply the ANNs for any petroleum application was presented and accomplished by a flowchart that can serve as a practical reference to apply the ANNs for any petroleum application. The method was broken down into steps that can be followed easily. The availability of huge data sets in the petroleum industry gives the opportunity to use these data to make better decisions and predict future outcomes. This paper will provide a review of applications of ANNs in petroleum engineering as well as a clear methodology on how to apply the ANNs for any petroleum application.
An operator decided to fracture three wells near an old one, but there was a fault in the vicinity crossing the path of the three laterals. The company decided to go ahead with the fracture anyway to see what would happen. It got an unusually detailed look. The fluid injected during a zipper fracture of three Anadarko Basin wells that flanked an older parent well was monitored with an electromagnetic imaging system developed by Deep Imaging. A technical paper presented Tuesday at the SPE Hydraulic Fracturing Technology Conference (SPE 194313) described how the fluid was illuminated by an electromagnetic signal from two transmitter lines placed on the surface parallel to the laterals.
The objective was to leverage prestack and poststack seismic data in order to reconstruct 3D images of thin, discontinuous, oil-filled packstone pay facies of the Upper and Lower Wolfcamp formation (Sakmarian time: 293-296 Ma).
The well-to-seismic tie was carefully established using synthetic seismograms, which enabled the facies log to be properly associated with the corresponding seismic samples. The seismic data were all resampled from 2 ms to 0.5 ms in anticipation of being able to recover facies thicknesses on the order of 2 m. Six neural networks with diverse learning strategies were trained to recognize the nine facies classes in the high-resolution seismic stack: Instantaneous Frequency, Instantaneous Q Factor, Inversion (P-Impedance), Semblance, Dominant Frequency, Most Negative Curvature, and eight Angle Stacks, using a two-stage learning and voting process.
At the wells, the nine facies were reconstructed from seismic at a 97% accuracy rate. The bootstrap classification rate, a proxy for blind well testing, was over 80%, which indicates a high-quality modeling process. The pay facies was described with no false positives or false negatives. In the 3D seismic volume between the wells, the procedure produced a Most Likely Facies volume (unsmoothed and smoothed), and nine individual Facies Probability volumes. The pay facies was visualized in a 3D voxel visualization canvas using opacity, and also in a two-way time thickness map. The usable vertical and horizontal resolution was much greater than that of the original seismic. Based on these classification results, additional drilling locations were chosen to further target the oil-filled packstones.
The classification results were created by neural networks, which can be used as a substitute for traditional AVO, inversion and cross-plotting techniques for seismic reservoir characterization. The time need to create the Machine Learning results for this small dataset was on the order of ten minutes.
Azevedo, Leonardo (Cerena/Decivil, Instituto Superior Técnico) | Demyanov, Vasily (Institute of Petroleum Engineering, Heriot-Watt University) | Lopes, Diogo (Cerena/Decivil, Instituto Superior Técnico) | Soares, Amílcar (Cerena/Decivil, Instituto Superior Técnico) | Guerreiro, Luis (Partex Oil & Gas)
Geostatistical seismic inversion uses stochastic sequential simulation and co-simulation as the perturbation techniques to generate and perturb elastic models. These inversion methods allow retrieve high-resolution inverse models and assess the spatial uncertainty of the inverted properties. However, they assume a given number of a priori parametrization often considered known and certain, which is exactly reproduce in the final inverted models. This is the case of the top and base of main seismic units to which regional variogram models and histrograms are assigned. Nevertheless, the amount of existing well-log data (i.e., direct measurements) of the property to be inverted if often not enough to model variograms and its histograms are biased towards the more sand-prone facies. This work shows a consistent stochastic framework that allows to quantify uncertainties on these parameters which are associated with large-scale geological features. We couple stochastic adaptive sampling (i.e., particle swarm optimization) with global stochastic inversion to infer three-dimensional acoustic impedance from existing seismic reflection data. Key uncertain geological parameters are first identified, and reliable a priori distributions inferred from geological knowledge are assigned to each parameter. The type and shape of each distribution reflects the level of knowledge about this parameter. Then, particle swarm optimization is integrated as part of an iterative geostatistical seismic inversion methodology and these parameters are optimized along with the spatial distribution of acoustic impedance. At the end of the iterative procedure, we retrieve the best-fit inverse model of acoustic impedance along with the most probable value for the location of top and base of each seismic unit, the most likely histogram and variogram model per zone. We couple stochastic adaptive sampling (i.e., particle swarm optimization) with global stochastic inversion to infer three-dimensional acoustic impedance from existing seismic reflection data. Key uncertain geological parameters are first identified, and reliable a priori distributions of potential values are assigned to each parameter. The type and shape of each distribution reflects the level of knowledge about this parameter. Then, particle swarm optimization is integrated as part of an iterative geostatistical seismic inversion methodology and these parameters are optimized along with the spatial distribution of acoustic impedance. At the end of the iterative procedure we retrieve the best-fit inverse model of acoustic impedance along with the most probable value for the location of top and base of each seismic unit, the most likely histogram and variogram model per zone.