Total Organic Carbon Prediction Research Articles

Seismic reservoir characterization of Utica-Point Pleasant shale with efforts at fracability evaluation — Part 2: A case study

The Utica Formation in eastern Ohio possesses all the prerequisites for being a successful unconventional play. Attempts at seismic reservoir characterization of the Utica Formation have been discussed in part 1, in which, after providing the geologic background of the area of study, the preconditioning of prestack seismic data, well-log correlation, and building of robust low-frequency models for prestack simultaneous impedance inversion were explained. All these efforts were aimed at identification of sweet spots in the Utica Formation in terms of organic richness as well as brittleness. We elaborate on some aspects of that exercise, such as the challenges we faced in the determination of the total organic carbon (TOC) volume and computation of brittleness indices based on mineralogical and geomechanical considerations. The prediction of TOC in the Utica play using a methodology, in which limited seismic as well as well-log data are available, is demonstrated first. Thereafter, knowing the nonexistence of the universally accepted indicator of brittleness, mechanical along with mineralogical attempts to extract the brittleness information for the Utica play are discussed. Although an attempt is made to determine brittleness from mechanical rock-physics parameters (Young’s modulus and Poisson’s ratio) derived from seismic data, the available X-ray diffraction data and regional petrophysical modeling make it possible to determine the brittleness index based on mineralogical data and thereafter be derived from seismic data.

A new method for TOC estimation in tight shale gas reservoirs

Total organic carbon (TOC) estimation is significantly crucial for shale reservoir characterization. Traditional TOC estimation methods (such as Passey and Schmoker method) do not provide accurate TOC predictions in shale gas reservoirs especially for the self-generated and self-stored reservoirs. This study proposes, for the first time, a new TOC prediction method based on Gaussian Process Regression (GPR) bridging geostatistics and machine learning technique. The method utilizes a non-parametric regression approach in shale TOC predictions, and not only provides the expert solutions in high-dimension processing, small samples and non-linear problems, but also has a better adaptation and generalization ability compared with other machine learning methods. The approach accounts for all the well logging attributes and chooses the relevant logs to build TOC estimation model, and 7 different kernel functions and 5 attributes groups are analyzed to get the optimized hyperparameters in practice. Application of the developed model to two shale gas reservoirs showed that the model predicted TOC matched well with that from the laboratory measurements. The proposed model based on GPR method provides an accurate way for the TOC prediction in the tight shale gas reservoirs.

Analyzing organic richness of source rocks from well log data by using SVM and ANN classifiers: A case study from the Kazhdumi formation, the Persian Gulf basin, offshore Iran

Determination of TOC is critical to the evaluation of every source rock unit. Methods which are dependent upon extensive laboratory testing are limited by the availability and integrity of the rock samples. Prediction of TOC (Total Organic Carbon) from well Log data being available for the majority of wells being drilled provides rapid evaluation of organic content, producing a continuous record while eliminating sampling issues. Therefore, the ideal method for determining the TOC fraction within source rock units would utilize common well log data. So a model was developed to formulate TOC values in the absence of laboratory TOC measurements from conventional well log data. Consequently, with the assistance of FL (Fuzzy Logic), TOC estimated from well log data with an overall prediction accuracy of 0.9425 for the test set. Following that TOC content of the Kazhdumi formation optimally has been divided into 4 zones using K-means cluster analysis, since searching for patterns is one of the main goals in data mining. There is a general increase in TOC from zone 1 to zone 4. The optimal number of zones has been detected by means of the knee method that finds the “knee” in a number of clusters vs. Compactness, Davies-Bouldin and Silhouette values. In the last step, using SVM (Support Vector Machine) and ANN (Artificial Neural Network) algorithms, two commonly used techniques, classification rules developed to predict the source rock class-membership (zones) from well log data. The proposed method is found effective in directly extracting patterns from well log data after defining classification rules. Quantitative comparisons of the results from ANN and SVM depicts that for classification problem of source rock zonation SVM with RBF (Radial Basis Function) kernel readily outperforms ANN in term of classification accuracy (0.9077 and 0.9369 for ANN and SVM, respectively), reduced computational time and highly repeatable results. This method would enable a more elaborate assessment of Kazhdumi formation to be undertaken by providing a comprehensive quick look results derived directly from well log data while using conventional methods one can’t define patterns within the data without grouping data manually.

Improving prediction of Total Organic Carbon in prospective Australian basins by employing machine learning

Total organic carbon (TOC) is directly associated with total porosity and gas content and is a critical factor in assessing the potential of unconventional reservoirs. TOC content is only known at the depths where the laboratory measurements on recovered core samples are performed. However, reliable estimation of potential resources can only be based on information about vertical and lateral distribution of organic matter throughout the prospective gas shale reservoir. This information is commonly obtained from conventional wireline logs, such as gamma ray, density, transit time and resistivity. Due to the complexity of unconventional reservoirs, traditional methods based on distinct differences of resistivity, density and sonic velocity of organic matter from those of the inorganic matrix are not always successful. We investigate the best way to predict the TOC using gamma-ray, density, porosity, resistivity and sonic transit time log responses by applying machine learning methods such as Artificial Neural Network (ANN) and Support Vector Machine (SVM). The analysis is done on the data from seven wells drilled through onshore unconventional reservoirs in the McArthur Basin (Northern Territory) and Georgina Basin (Northern Territory and Queensland), Australia. The prediction quality of traditional, multiple liner regression (MLR) and machine learning methods was compared. The most accurate TOC estimates were generated by ANN- and SVM-based nonlinear predictors, followed by the MLR and traditional models. This indicates that geologic complexity affects the relationship between the log response and TOC in the area of interest.

Application of extreme learning machine and neural networks in total organic carbon content prediction in organic shale with wire line logs

Total organic carbon (TOC) is a critical parameter for source rock characterization in shale gas reservoirs. In this work, the use of extreme learning machines (ELM) for predicting TOC from well logs data have been investigated. We use log data from two wells located in an unconventional shale gas reservoir in the Sichuan Basin, China. Seven wireline logs from this well and a total of 185 TOC observations from core measurements were incorporated. Prediction accuracy of the model has been evaluated and compared with commonly used artificial neural network which is based on Levenberg-Marquardt logarithm (ANN-LM). An Extreme Learning Machine (ELM) network is a single hidden-layer feed-forward network with many advantages over multi-layer networks, such as fast computing speed and better generalization performance. The results demonstrated that TOC prediction by the ELM model and the ANN model, but the ELM method can achieve high accuracy while maintains high running speed. This study shows that ELM technology is a promising tool for TOC prediction, and this work can be incorporated into a software system that can be used in quick ‘sweet spot’ determination and well completion guidance.

Selecting optimum log measurements for hydraulic fracturing

We have developed a statistical method for investigating the importance of different log measurements for picking the best zones for hydraulic fracturing. We have determined the method’s applicability using data from unconventional reservoirs (Eagle Ford, Haynesville, Barnett, and a reservoir from the Middle East). The analysis began with single log measurements (e.g., gamma ray [GR], compressional and shear sonic [DTC and DTS], and spectral gamma ray [SGR], which could measure the radioactivity of uranium [U], potassium [K], and thorium [Th]). Other types of measurements, including density (RhoB), neutron porosity (NPHI), and resistivity, were added to obtain more complex logging suites. These log measurements were the inputs for this analysis. Each input combination was referred to as a “scenario.” Parameters such as effective porosity (PhiE), brittleness, total organic carbon (TOC), production index (PI), and fracture index (FI) were referred to as the outputs for the analysis. We have investigated linear and nonlinear combinations of the inputs to predict the outputs. Various scenarios, beginning with the simplest cases and ending with the most complete combination, were tested. The selection of log combinations was either based on the importance of individual logs or on industry-standard combinations (such as triple and quad combos). For each scenario, we computed correlation coefficients and root-mean-square errors of predicting the output parameters. The prediction accuracies generally increased as a result of increasing the number of input logs. Our analysis clearly found the importance of using SGR (for PI and FI prediction) and resistivity (for TOC prediction) logs. Based on comparison of the reconstruction results, actual values, and correlation coefficients/errors, we ranked the log combinations for predicting/modeling a specific parameter. The most challenging properties to model included TOC, PhiE, PI, and FI; the easiest properties to predict were brittleness and Young’s modulus.

Revised models for determining TOC in shale play: Example from Devonian Duvernay Shale, Western Canada Sedimentary Basin

Determination of total organic carbon (TOC) is essential in unconventional shale resource play evaluation. Indirect method, such as petrophysical approach, can provide a fast, convenient and cost efffective means for TOC estimation from well logs. Among the publically available approaches, the ΔlogR method is a popular one in conventional source rock evaluation and has been applied to unconventional resource play evaluation by many practitioners. Careful examination of the method finds that improvements can be made for a better TOC estimation with relaxed limitations. This study proposes the following revisions on the original ΔlogR method: 1) allowing estimation of petrophysical parameters from the targeted source rock rather an assumed linear approximation, 2) using additional logs to improve TOC prediction, and 3) replacing LOM (level of organic metamorphism unit) with a more commonly used thermal indicator Tmax for convenience. Depending on data or preference, the choice of using different porosity logs (density versus sonic) and parameter estimation methods (cross-plot method versus data-driven approaches) makes the revised method flexible for application. We present the revisions along with application examples from the Devonian Duvernay Shale in the Western Canada Sedimentary Basin to demonstrate the application of the improved ΔlogR method in shale gas evaluation. Comparison of the results from the revised method and the original one reveals that the revised models provide better and unbiased estimates of TOC with a higher correlation coefficient and bell-shaped residues centered at zero.

Support-vector-regression machine technology for total organic carbon content prediction from wireline logs in organic shale: A comparative study

Organic shale is one of the most important unconventional oil and gas resources. Hydrocarbon potential prediction of organic shale such as total organic carbon (TOC) is an important evaluation tool, which primarily uses empirical equations. A support-vector machine is a set of supervised tools used for classification and regression problems. In this study, a support-vector machine for regression (SVR) is investigated to estimate the TOC content in gas-bearing shale. First, SVR technology is introduced including its basic concepts, associated regression algorithms and kernel functions, and a TOC prediction sketch that uses wireline logs. Then, one example is considered to compare three different regression algorithms and four different kernel functions in a packet dataset validation process and a leave-one-out cross-validation process. Error analysis indicates that the SVR method with the Epsilon-SVR regression algorithm and the Gaussian kernel produces the best results. The method of choosing the optimum Gamma value in the Gaussian kernel function is also introduced. Next, for comparison, the SVR-derived TOC with the optimal model and parameters is compared with the empirical formula and the ΔlogR methods. Finally, in a real continuous TOC prediction using wireline logs, TOC prediction tests are performed using SVR to choose the optimal logs as inputs, and the optimal input is finally chosen. Additionally, the radial basis network (RBF) is also applied to perform tests with different inputs; the results of these tests are compared with those of the SVR method. This study shows that SVR technology is a powerful tool for TOC prediction and is more effective and applicable than a single empirical model, ΔlogR and some network methods.

Total Organic Carbon Prediction in Shale Gas Reservoirs from Well Logs Data Using the Multilayer Perceptron Neural Network with Levenberg Marquardt Training Algorithm: Application to Barnett Shale

The main goal of this paper was to predict the total organic carbon (TOC) from well logs data using the multilayer perceptron (MLP) neural network machine. This can replace the Schmoker’s method in case of discontinuous measurement of the bulk density log. The MLP machine is composed of three layers, an input layer with four neurons corresponding to the Gamma ray, the neutrons porosity, and the slowness of the P and S waves well logs. The output layer is formed with one neuron, which corresponds to the predicted TOC log, and a hidden layer with ten neurons. The MLP machine is trained using the Levenberg–Marquardt algorithm. Data of two horizontal wells drilled in the lower Barnett formation located in USA are used. Comparison between the predicted and calculated TOC log using the Schmoker’s method clearly shows the use of the neural network method to predict the TOC in shale gas reservoirs.

A dynamic adaptive radial basis function approach for total organic carbon content prediction in organic shale

Total organic carbon (TOC) is an important parameter for characterizing shale gas and oil reservoirs. Estimation of TOC from well logs has previously been achieved by an empirical model. The radial basis function (RBF) neural network is a new quantitative method that can generate a smooth and continuous function of several input variables to approximate the unknown forward model. We investigated the basic principles of the RBF including network structure, basis function, network training method, and its application in the TOC prediction. The nearest neighbor algorithm was selected for the network training. Then, the Gaussian width was investigated to improve the TOC prediction accuracy through leave-one-out cross-validation. Finally, field cases of organic shale were studied for the TOC prediction, and the prediction results using the RBF method were compared with those of the [Formula: see text] method. Furthermore, according to sensitive attribute ranking, the impacts of different input logs on the predicted results were also investigated through various experiments, and the best network model was finally chosen. The error analysis between the prediction results and lab-measured TOC in some examples indicated that the new approach is more accurate than a single empirical regression method and more flexible than the [Formula: see text] method.

Predicting Total Organic Carbon Removal Efficiency and Coagulation Dosage Using Artificial Neural Networks

Total organic carbon (TOC) in raw water is considered one of the major precursors to formation of disinfection by-products. Many drinking water treatment plants in the United States depend on coagulation, flocculation, and precipitation processes to bring treated water TOC levels to within acceptable limits. However, predicting TOC removal efficiency for a known coagulant dosage is not a trivial task and requires popular but time-consuming jar tests. This article presents successful implementation of artificial neural network (ANN) technology in predicting the TOC removal efficiency based on routinely measured physical and chemical raw-water characteristics and coagulant dosage. ANN predictions were better than multiple linear regression models. A new promising approach to develop a dose-response curve from the trained ANN is proposed and demonstrated in this study. This approach will facilitate instantaneous prediction of TOC at the outlet. It can also help in estimating optimal coagulant dosage in a drinking water treatment plant.

Comparison of intelligent and statistical clustering approaches to predicting total organic carbon using intelligent systems

We propose a two-step approach in predicting Total Organic Carbon (TOC) content from well log data. Initially, the well log data are classified into a set of electrofacies (EF). This classification does not require any further subdivision of the dataset but follows naturally based on the unique characteristics of well log measurements reflecting mineral and lithofacies responses within the logged intervals. In this study, the Self-Organizing Maps (SOM) as the intelligent data clustering methods are compared with the statistical clustering approaches including the Hierarchical Cluster Analysis (HCA) and K-means clustering to characterize and identify electrofacies. The results obtained from the all methods are compared to each other and the best method is chosen based on the cluster validity tests to clustering the petrophysical data into a certain number of EF. Afterwards, the TOC values are estimated from well log data by using the individual Artificial Neural Network (ANN) models constructed for each EF.In the second approach, the TOC data are estimated for the total interval by using a similar ANN model regardless of data clustering and EF determination. The results of two prediction methods are compared to each other and a third conventional ∆ log R technique as well. The results show that clustering of a formation into specific units (electrofacies) provides better results in TOC prediction compared to the models constructed for the whole dataset as a single cluster. In addition, intelligent systems are more efficient than the previous conventional techniques based on ∆ log R method. The proposed methodology is illustrated using a case study from the world's largest non-associated gas reservoir, the South Pars Gas Field, the Persian Gulf basin.

The prediction of soil carbon fractions using mid-infrared-partial least square analysis

This paper describes the application of mid-infrared (MIR) spectroscopy and partial least-squares (PLS) analysis to predict the concentration of organic carbon fractions present in soil. The PLS calibrations were derived from a standard set of soils that had been analysed for total organic carbon (TOC), particulate organic carbon (POC), and charcoal carbon (char-C) using physical and chemical means. PLS calibration models from this standard set of soils allowed the prediction of TOC, POC, and char-C fractions with a coefficient of determination (R2) of measured v. predicted data ranging between 0.97 and 0.73. For the POC fraction, the coefficient of determination could be improved (R2 = 0.94) through the use of local calibration sets. The capacity to estimate soil fractions such as char-C rapidly and inexpensively makes this approach highly attractive for studies where large numbers of analyses are required. Inclusion of a set of soils from Kenya demonstrated the robustness of the method for total organic carbon and charcoal carbon prediction.