Placing horizontal wells in the correct zones of the producing reservoir in static/dynamic models is important for robust model quality and essential for production history matching. A laborious technique of manually generating correction points around each un-calibrated well is often used by geomodelers. This technique is not systematic and is highly interpretive. We present an automated global solution that leverages not only the well tops, but also the well trajectory information to calibrate the horizontal wells. The solution comprises a series of processes that can all be run through a custom built graphical user interface (GUI). The processes are designed to a) detect the calibration problems with the ability to visualize them, b) correct each surface separately for zone mismatch and c) retrieve original zone thickness whenever possible. Treating each surface separately simplifies the problem and causes great reduction in run time compared to simultaneous correction of all surfaces in one go. However, the final results are essentially global, which is insured by maintaining the input zone thicknesses (except if well data confirms otherwise). The correction is achieved by ensuring that model zones (between the input surfaces) match the "optimum zone logs" at all wells in the model. The term optimum, in this context, means that the zone log should only contain the zones to be modeled, no more no less. These optimum zone logs are typically calculated from the well tops, which implies having an impeccable set of well tops. In most cases, well tops sets may contain extra tops (that are not included in the model) and missing tops at some of the wells. While the latter problem is more difficult to solve, both problems must be detected and corrected for a successful run of this solution. Fortunately, the "diagnosis" step at the beginning of this solution detects all these problems, which reduces the time needed to review the well correlation, especially in fields with hundreds of wells. Moreover, machine learning techniques are used to suggest the position of the missing well tops. The user gets to either approve the suggestion or change it manually. The solution we are presenting is fully automatic and fully reproducible. It is given with four parameters to control the amount and influence of the correction applied to the horizon intersecting the trajectory when it is not supposed to. If optimum zone logs are ready, the solution can run in few minutes to correct large models with hundreds of wells and tens of surfaces. This has the potential of reducing months of work to few days to calibrate horizontal wells in a structural model. The few days may include optimization of the solution parameters to achieve the desired results.

Effective well placement and design planning accounts for subsurface uncertainties to estimate production and economic outcomes. Reservoir modelling and simulation workflows build on ensemble approaches to manage uncertainties for production forecasting. Ensemble generation and interpretation requires a higher degree of automation analytics and artificial intelligence for fast value extraction and decision support. This work develops practical intelligent workflow steps for a robust infill well placement and design scenario in multi-layered/stacked reservoirs under uncertainty. Potential well targets are classified by an opportunity index defined by a combination of rock and hydrocarbon flow properties as well as connected volumes above a minimum economic volume. Unsupervised learning techniques are applied to automate the search for alternative target areas, so-called hotspot regions. Supervised machine/learning models are used to predict infill well performance based on simulated and/or past production experience. A stochastic evaluation including all ensemble cases is used to capture uncertainty. Vertical, deviated, horizontal and multilateral wells are proposed to optimally target single or connect to multiple hotspot regions under technical and economic constraints. A structured workflow design is applied to a multi-layered/stacked reservoir model. Subsurface uncertainties are described and captured by multiple model realizations, which are constrained in areas of historical wells. An infill well program for a multi-layered/stacked reservoir is defined for incremental production increase under economic constraints. This work shows how robust well location and design builds on the full ensemble of cases with a high degree of automation using analytics and machine-learning techniques. Both production and economic targets are calculated and compared to a reference case for robust solution verification and probability of success. In conclusion, an overall reservoir-driven field development strategy is required for efficient execution. However, automation is well applicable to repetitive workflow steps which includes hotspot search in an ensemble of validated reservoir models. This work presents an integrated, intelligent solution for informed decision making on infill drilling locations and refined well design. Higher degree of automation with embedded intelligence are discussed from case generation to hotspot identification. Aspects of model calibration in a producing field environment are addressed.

Naturally fractured carbonate reservoirs (NFR) host 50% of the world hydrocarbon reserves. Carbonate reservoirs are known for their high degree of heterogeneity and uncertainty in reservoir description. Characterizing the fracture system with reduced uncertainty helps in building predictable reservoir models that in turn are used by reservoir management for business decisions. Integration of multi-data sources (static and dynamic) is vital to the understanding of the mechanisms of fluid flow present in a given reservoir. Calibration of geologic-based models (conditioned by static data) to flow-related data (well test and production data) can dramatically reduce the uncertainty in reservoir models. In this work, we present a new development to further reduce the uncertainty in the characterization of fracture properties (e.g., orientation, conductivity, aperture, length and density) from well test pressure responses (e.g., permeability-thickness product, storativity, and interporosity). The optimization problem is addressed using a direct search method. A novel multi-level genetic algorithm is developed to find the optimum solution space of the fracture properties by minimizing the error in a new multi-objective function. The proposed algorithm was benchmarked against the industrial software FracaFlow©. Synthetic data of heterogeneous systems were used for validation as well as to demonstrate the new algorithm capabilities. Our results clearly show further reduction of uncertainties in fracture property estimation compared to FracaFlow©.

Integrated, multidisciplinary field projects require the transfer of large datasets and results from one discipline to another. Typically, petrophysical workflows for geological modeling produce only rock-types, porosity, permeability, and water saturation model logs. Petrophysical logs are upscaled to the fine geological model grid during the 3D modeling workflow. The geological model aerial cell dimensions and layer size are determined using geostatistical methods to optimize the run time during the dynamic simulation phase. In this study, we propose to enhance the number of outputs from the petrophysical workflow by providing recommendations on the optimal vertical layer size by stratigraphic interval for capturing all relevant heterogeneities associated with complex carbonate reservoirs. We assume that the deflections of the petrophysical logs (such as porosity or permeability) are symmetrical across the geological layer. This is a reasonable assumption for vertical and subvertical wells since the layer size should be comparable to the vertical resolution of the logs. Therefore, the inflexion points of the input log correspond to the facies changes. After identifying the layers’ borders, we calculate the thickness of each layer and its corresponding input log value. The developed application was successfully tested using data from more than 200 wells of a giant carbonate reservoir from the Middle East. The results demonstrate that the developed algorithm parameters are easily adaptable to a variety of both different input log types and various ranges of petrophysical properties. The created interface interactivity enables the user to readily assess and alter parameters to optimize the layering scheme while capturing the right level of heterogeneities. During the testing on a real dataset, the developed application revealed vital information on the formation's vertical heterogeneity and the significant difference between distinct stratigraphic intervals. The provided analysis of layer thicknesses can be used directly for the creation of static model layering scheme.

Permeability modelling remains a major challenge in the reservoir modelling exercise. The main reason for this is the limited availability of measured input data and the effect of different geological processes on reservoir permeability. This leads to nonrepresentation of high-permeability streaks in the model. In this paper, we present a machine-learning (ML) driven approach that captures the permeability variation in the reservoir using available input data. In ML, clustering is an unsupervised approach aimed at automatically grouping data with similar properties. We use several clustering techniques to automatically identify high-permeability data points by dividing data into groups, also known as clusters, and then choosing the cluster with the maximum permeability and assigning it a new rock type. For each rock type, we fit and evaluate many ML regression models, and show their outperformance over traditional fitting approaches. Porosity and several openhole log properties are used as input for the regression models. By fixing porosity but varying the other properties, the variability of permeability values is predicted. Clustering, using ‘K-Means’ ML algorithm, resulted in an efficient approach of automated high permeability identification. Several ML models were trained and evaluated, and the models with the minimum error scores, namely mean square error (MSE) and R Squared (R2), were chosen for further predictions. Random Forest was within the top models for a variety of rock types. In general, complex curve fitting using ML outperformed traditional fitting approaches (i.e., straight line fitting) and demonstrated high potential for accurate, automated, high-permeability identification and integration. The predicted permeability has been calibrated with well test permeability data.