Reservoir Characterization Using Intelligent Seismic Inversion

Journal of Petroleum Exploration and Production, 2021

In an attempt to reduce the errors and uncertainties associated with predicting reservoir properties for static modeling, seismic inversion was integrated with artificial neural network for improved porosity and water saturation prediction in the undrilled prospective area of the study field, where hydrocarbon presence had been confirmed. Two supervised neural network techniques (MLFN and PNN) were adopted in the feasibility study performed to predict reservoir properties, using P-impedance volumes generated from model-based inversion process as the major secondary constraint parameter. Results of the feasibility study for predicted porosity with PNN gave a better result than MLFN, when correlated with well porosity, with a correlation coefficient of 0.96 and 0.69, respectively. Validation of the prediction revealed a cross-validation correlation of 0.88 and 0.26, respectively, for both techniques, when a random transfer function derived from a given well is applied on other well lo...

Computers & Geosciences, 2011

In this work, we address the problem of transforming seismic reflection data into an intrinsic rock property model. Specifically, we present an application of a methodology that allows interpreters to obtain effective porosity 3D maps from post-stack 3D seismic amplitude data, using measured density and sonic well log data as constraints. In this methodology, a 3D acoustic impedance model is calculated from seismic reflection amplitudes by applying an L 1 -norm sparse-spike inversion algorithm in the time domain, followed by a recursive inversion performed in the frequency domain. A 3D low-frequency impedance model is estimated by kriging interpolation of impedance values calculated from well log data. This low-frequency model is added to the inversion result which otherwise provides only a relative numerical scale. To convert acoustic impedance into a single reservoir property, a feed-forward Neural Network is trained, validated and tested using gamma-ray and acoustic impedance values observed at the well log positions as input and effective porosity values as target. The trained NN is then applied for the whole reservoir volume in order to obtain a 3D effective porosity model. While the particular conclusions drawn from the results obtained in this work cannot be generalized, such results suggest that this workflow can be applied successfully as an aid in reservoir characterization, especially when there is a strong non-linear relationship between effective porosity and acoustic impedance.

Abu Dhabi International Conference and Exhibition, 2004

TX 75083-3836, U.S.A., fax 01-972-952-9435.

2002

The project, "Calibration of Seismic Attributes for Reservoir Calibration," has completed the initially scheduled third year of the contract, and is beginning a fourth year at no additional cost, designed to expand upon the tech transfer aspects of the project. From the Stratton data set we have demonstrated that an apparent correlation between attributes derived along 'phantom' horizons are artifacts of isopach changes; only if the interpreter understands that the interpretation is based on this correlation with bed thickening or thinning, can reliable interpretations of channel horizons and facies be made. From the Boonsville data set we developed techniques to use conventional seismic attributes, including seismic facies generated under various neural network procedures, to subdivide regional facies determined from logs into productive and non-productive subfacies, and we developed a method involving crosscorrelation of seismic waveforms to provide a reliable map of the various facies present in the area. The Teal South data set has provided a surprising set of data, leading us to develop a pressure-dependent velocity relationship and to conclude that nearby reservoirs are undergoing a pressure drop in response to the production of the main reservoir, implying that oil is being lost through their spill points, never to be produced. The Wamsutter data set led to the use of unconventional attributes including lateral incoherence and horizon-dependent impedance variations to indicate regions of former sand bars and current high pressure, respectively, and to evaluation of various upscaling routines.

In field development and production, a reservoir model is a key element in the successful performance of an oil and gas field. Well logs and core data have high vertical resolution whereas seismic data has poorer resolution vertically. However, 3D seismic data has very high horizontal resolution. We will take this into account while proposing the new methodology. Depending on wells only we can have a hires model far higher than what we actually need. However, in between the wells any meaningful extrapolation is flawed. Using geo-statistics through kriging algorithm and variography leads to non-geologic reservoir models. One particular process which is highly undesirable is to throw any valid well info and upscale it such as to fit the seismic. This results in a model that is inherently flawed. The important aspect of the new methodology is to follow up the seismic stochastic inversion together with a suitable rock physics modelling to achieve high resolution (reservoir scale) of different litho-facies discrimination. In this study first a robust rock physics model is performed for not only shear velocity log prediction at not having shear log wells but also to more appropriate litho-facies differentiation at well locations using seismic elastic properties (AI vs. Vp/Vs). Afterwards, seismic pre-stack stochastic inversion is carried out to populate different types of litho-facies between the wells. Eventually, the distribution map of pay zone facies is resulted.

A supervised seismic facies analysis based on artificial neural network was carried out in order to characterize a vuggy carbonate reservoir located in Barinas, Southwest of Venezuela. With this purpose a neural network was training using a back-propagation algorithm on values of petrophysical attributes obtained from resistivity and acoustic image logs, as well as values of seismic attributes close to the wells at which those reservoir properties were measured. The seismic attributes used for the classification were obtained after applied principal component analysis to a set of seismic attributes, which included complex trace attributes and acoustic impedance values obtained from a seismic inversion. As a result, it was generated a seismic facies map that extrapolates information of porosity obtained from borehole image logs and characterises the heterogeneity of the reservoir.

Journal of Applied Geophysics

Velocity deviation log (VDL) is a synthetic log used to determine pore types in reservoir rocks based on a combination of the sonic log with neutron-density logs. The current study proposes a two steps approach to create a map of porosity and pore type by integrating the results of petrographic studies, well logs and seismic data. In the first step, velocity deviation log was created from the combination of the sonic log with the neutron-density log. The results allowed identifying negative, zero and positive deviations based on the created synthetic velocity log. Negative velocity deviations (below-500 m/s) indicate connected or interconnected pores and fractures, while positive deviations (above +500 m/s) are related to isolated pores. Zero deviations in the range of [-500m/s, +500m/s] are in good agreement with intercrystalline and microporosities. The results of petrographic studies were used to validate the main pore type derived from velocity deviation log. In the next step, velocity deviation log was estimated from seismic data by using a probabilistic neural network model. For this purpose, the inverted acoustic impedance along with the amplitude based seismic attributes were formulated to VDL.

EAGE 63rd Conference & Technical Exhibition — Amsterdam, 2001

The aim of any reservoir characterization study is to describe the properties of the reservoir in the appropriate detail. The result is a prediction model of the physical and geological properties of the subsurface. To reach this goal we must integrate spatially distributed seismic data measured in two way time and local informations from wells measured in units of depth. To handle these data of different character and scale a neural network training approach was selected to provide various attribute maps without well (unsupervised) and with well control (supervised) finally leading to volumes of distributed properties as lithology, porosity and porevolumes. The procedure of distributing well properties to 3D space, called volume transformation, can be repeated for newly acquired or modeled 3D seismic data such leading to a tool for monitoring or predicting the properties of the reservoir. In this study we use a 3D seismic volume with an interpreted horizon and 12 wells located in the area with the following logs: sonic, density, porosity and lithofacies. The paper describes a case study of quantitative interpretation using modelled log and seismic data in order to simulate and predict the responses of a gas storage reservoir in the Paris Basin.

NRIAG Journal of Astronomy and Geophysics

The anisotropic and heterogeneous nature of the subsurface units at offset wells gets more complex at locations outside well control. However, qualitative and quantitative predictions of reservoir properties and geometries beyond well control are vital to understanding the intrinsic characteristics of subsurface formations. Supervised Multi-layer Perceptron neural network and conditional sequential Gaussian simulation were applied to a suite of well logs data set and three dimensional (3-D) seismic dataset acquired from P-field, Niger Delta to predict lateral continuity of hydrocarbon reservoir properties. Multilayer perceptron neural network was used to model effective porosity (Eϕ) at the root-mean-square error of 0.00531 with an estimated average effective porosity of 0.2952. Furthermore, MLPNN modelling of hydrocarbon saturation (S h) at root mean square error of 0.02821 yielded an estimated average of 69.73%. The volume of shale (V sh) was also modelled at the root mean square error of 0.0282, with an estimated Vsh average of 9% for the study area. It was discovered that a comparatively higher net-to-gross (N-T-G), relatively higher effective porosity, high hydrocarbon saturation, very low volume of shale, and high permeability were observed in the areas of interest in the study area. A geostatistical approach was also used to model petrophysical properties. The parametric semivariogram model shows the range of 94-5230 m, nugget effect of 0.062, and sills of 0.075, 0.093, and 0.0121. Realisations were generated and ranked using the SGS algorithm suggest that any one of the realisations can independently represent the real picture of the subsurface geology within the study area. The integration of these different prediction tools and analyses of the outcomes from the research has improved our understanding of delineated reservoirs and improved lateral variation prediction of its properties. These prediction tools served better as a complementary tool to each other rather than as a comparison tool.

Handbook of Mathematical Geosciences, 2018

In the oil industry, exploratory targets tend to be increasingly complex and located deeper and deeper offshore. The usual absence of well data and the increase in the quality of the geophysical data, verified in the last decades, make these data unavoidable for the practice of oil reservoir modeling and characterization. In fact the integration of geophysical data in the characterization of the subsurface petrophysical variables has been a priority target for geoscientists. Geostatistics has been a key discipline to provide a theoretical framework and corresponding practical tools to incorporate as much as possible different types of data for reservoir modeling and characterization, in particular the integration of well-log and seismic reflection data. Geostatistical seismic inversion techniques have been shown to be quite important and efficient tools to integrate simultaneously seismic reflection and well-log data for predicting and characterizing the subsurface lithofacies, and its petro-elastic properties, in hydrocarbon reservoirs. The first part of this chapter presents the state of the art and the most recent advances of geostatistical seismic inversion methods, to evaluate the reservoir properties through the acoustic, elastic and AVA seismic inversion methods with real case applications examples. In the second part we present a methodology based on seismic inversion to assess uncertainty and risk at early stages of exploration, characterized by the absence of well data for the entire region of interest. The concept of analog data is used to generate scenarios about the morphology of the geological units, distribution of acoustic properties and their spatial continuity. A real case study illustrates the this approach.