Autor: João Lucas Braga Da Silva

Evaluation of unsupervised machine learning frameworks to select representative geological realizations for uncertainty quantification

Publicado em por .

Ensembles of geological realizations (GR) are normally processed by numerical simulators to evaluate geological uncertainty during the decision-making process. Although different stochastic spatial algorithms can quickly generate hundreds to thousands of GR to capture the full uncertainty range, the simulation process applied to this number of realizations is computationally expensive. Hence, a small subset of representative geological realizations (RGR) that statistically represent the features of the full ensemble can be used for uncertainty quantification. In this study, unsupervised machine learning (UML) is applied by considering different (1) adjacency matrix construction, (2) dimensionality reduction, and (3) clustering and sampling algorithms to generate several RGR sets. Then, the mismatches between the distribution of different field and well indicators obtained from the RGR sets and the whole ensemble are measured using the Kolmogorov-Smirnov (KS) test to compare the uncertainty space of the subsets and the full set. Furthermore, to measure the pairwise adjacency between the realizations, we use a static reservoir feature called reservoir quality index (RQI). We performed extensive computational analyses to appraise the performance of the UML in two benchmark cases. Each case contains 500 GR. This study can provide a comprehensive assessment of the UML for the RGR selection due to the application of different algorithms. The results showed that the RGR set can be successfully selected without previous flow simulation runs, if an appropriate UML method is employed. This leads to a reduction in the computational cost during uncertainty quantification and risk analysis. Furthermore, we observed that the optimal number of RGR should be chosen due to the geological complexity of each case study. We also found that the type of recovery mechanism has no impact on the optimal number of RGR and on UML methods. The appropriate RGR set can be used for production forecasts and development planning support.

 

by Seyed Kourosh Mahjour, Luís Otávio Mendes da Silva, Luis Augusto Angelotti Meira, Guilherme Palermo Coelho, Antonio Alberto de Souza dos Santos, Denis José Schiozer, published at Journal of Petroleum Science and Engineering, February 2022, Vol. 209, 109822.

Fault and fracture study by incorporating borehole image logs and supervised neural network applied to the 3D seismic attributes: a case study of pre-salt carbonate reservoir, Santos Basin, Brazil

Publicado em por .

Fractures play a significant role in the development and production phases of carbonate reservoirs. Quantitative interpretation of fractures not only enhances reservoir models but also reduces the drilling risk and optimizes well design. In this study, we attempt to predict the fracture density map by integrating well and seismic data along with maximum horizontal stress identification. To this end, we propose a workflow with a set of machine learning approaches. First, 3D seismic data is conditioned after the migration processing sequence and the main faults and horizons are interpreted. Next, a number of curvature and coherence attributes are created for a supervised neural network technique to generate new seismic-based discontinuity attribute. Using a geostatistical method to incorporate the interpreted dip and azimuth attributes from well image logs and 3D seismic discontinuity attribute, the fracture density map is predicted and the results validated with a blind well. Finally, we evaluate the strike azimuth of possible open fractures based on the stress regime analysis, from which two distinctive zones are identified. There are, however, some limitations in this study. The predicted fracture density map can be employed to build a discrete fracture network, update dual porosity and permeability estimation, and identify sweet spots.

Experimental investigation of the Electrical Submersible Pump’s energy consumption under unstable and stable oil/water emulsions: A catastrophic phase inversion analysis

Publicado em por .

The presence of water in crude oil exploitation by the Electrical Submersible Pump (ESP) systems may cause several problems in energy consumption and operational instabilities due to emulsion formation. Indigenous surfactants in crude oil also contribute to emulsion stabilization, which can exacerbate these problems. In this paper will be experimentally investigated the influence of the emulsion stability on ESP energy consumption and operational instabilities through an 8-stage ESP operating with unstable and stable emulsions, with and without a demulsifier. The experimental tests were performed for one oil viscosity, a constant total flow rate, and two ESP rotational speeds. Initially, the ESP relative dimensionless power (RDP) was analyzed along with the emulsion system and the droplet size distribution (DSD). An interesting difference regarding the presence of surfactants was observed experimentally in the RDP and phase inversion points. The relationship among the ESP dimensionless power, torque, and electrical current with maximum droplet size allowed to conclude that these parameters can be related to the start of the coalescence process, i. e, able to predict the catastrophic phase inversion (CPI) point.

Scenario reduction methodologies under uncertainties for reservoir development purposes: distance-based clustering and metaheuristic algorithm

Publicado em por .

The simulation process under uncertainty needs numerous reservoir models that can be very time-consuming. Hence, selecting representative models (RMs) that show the uncertainty space of the full ensemble is required. In this work, we compare two scenario reduction techniques: (1) Distance-based Clustering with Simple Matching Coefficient (DCSMC) applied before the simulation process using reservoir static data, and (2) metaheuristic algorithm (RMFinder technique) applied after the simulation process using reservoir dynamic data. We use these two methods as samples to investigate the effect of static and dynamic data usage on the accuracy and rate of the scenario reduction process focusing field development purposes. In this work, a synthetic benchmark case named UNISIM-II-D considering the flow unit modelling is used. The results showed both scenario reduction methods are reliable in selecting the RMs from a specific production strategy. However, the obtained RMs from a defined strategy using the DCSMC method can be applied to other strategies preserving the representativeness of the models, while the role of the strategy types to select the RMs using the metaheuristic method is substantial so that each strategy has its own set of RMs. Due to the field development workflow in which the metaheuristic algorithm is used, the number of required flow simulation models and the computational time are greater than the workflow in which the DCSMC method is applied. Hence, it can be concluded that static reservoir data usage on the scenario reduction process can be more reliable during the field development phase.

A comparison of machine learning surrogate models for net value

Publicado em por .

Net Present Value (NPV) is an important indicator to guide investment decisions. In oil production planning, NPV is employed to evaluate and select among different production strategies. However, NPV estimation requires computational costly numerical simulations. So, evaluating as many production strategies as is desirable may be prohibitive. Therefore, one can only evaluate a small part of the search space, decreasing the chance of finding a near-optimal production strategy. To speed up the searching process, a much faster, but error-prone, surrogate model is used to approximate the simulator output. Data-driven surrogate modeling depends on both: 1) building a simple model to reproduce the quality of a high-fidelity model, while 2) considering a large volume of data to build it. In this work, we address the well placement optimization task by considering a binary data representation, indicating the presence or absence of a given well in a production strategy. We show the possibility of predicting the NPV from binary data, thus reducing data dimension and model complexity. Specifically, we compare six machine learning regression algorithms to predict the NPV. The simulations conducted in a benchmark case, based on a real field, showed that some regression algorithms can be used as a surrogate model to the simulator to efficiently perform well placement optimization considering binary data. The best results were obtained with Multi-Layer Perceptron, whose estimations covered a wide range of NPV with a small and constant error.