The oil & gas (O&G) industry is one of the most important activities that support the economy in the world. Reservoir production management is a challenge with several facets and, surely, one of great interest refers to the difficult forecasting of O&G production in a reservoir. Most models in the prior art are physics-based, and try to predict the reservoir behavior based on fluid dynamics simulation. The problem with these models is the high computational cost footprint as models can take several hours/days/weeks of computation to obtain an accurate simulation (Zhang, 2018). On the other hand, Machine Learning (ML) algorithms have been implemented in several applications, and they can lead to breakthroughs in several areas. Recently, they have been used in time series forecasting, where we try to predict a variable from historical records. It is very promising for O&G management as ML methods could be exploited for production forecasting. One advantage is the computational footprint of ML algorithms that are smaller as we can deploy most tasks in GPUs so highly parallelizing them. ML-based reservoir models have been classified into two major classes: first, the Surrogate Reservoir Models (SRM), in which the simulation is based on synthetic numerical models and try to reproduce accurate replicas of traditional reservoirs; and second, Top-down Models, when an ML model is built from actual field data such as historical production, seismic attributes, well production, etc. (Mohaghegh, 2011). The main challenge on all these models is to find the optimal input set that allows us to perform an accurate production forecasting. Sometimes it is difficult to find relationships, or correlations, between input and production variables. Common methods of production forecasting are based on regression algorithms such as Support Vector Machines (Noshi et al. 2019), Random Forests (Maucec and Garni, 2019), or Radial Basis Functions models (Memon et al. 2014). All these ML models share a similar methodology: first, they seek correlations between input values (i.e., injection wells, bottom-hole pressure, well logs, etc.); second, they define a training set from the input and the production variables; third, they train an ML model; and, finally, they validate the model with a testing set. If results from the regression model are close to the ones in the testing set, we can say the ML model has learned properly the mapping function between the input variables and the target variable, and the trained model can be used in forecasting. In this work, we present a forecasting model for O&G production based on a data-driven approach and supported by ML algorithms. Our model takes advantage of a long-short term memory (LSTM) methodology capable of finding correlations in the data not only in the recent data points of a time series but also more subtle ones present in longer time intervals. The results show promising perspectives for forecasting a short-term context for oil, gas, water, and liquid production on a synthetic, but realistic, benchmark.
Month: September 2020
Two-Stage Scenario Reduction Process for An Efficient Robust Optimization
Well-positions in an oil field have a key role in production performance and financial interests. Defining the location of wells is challenging due to rock-fluid interaction, adjacent wells effects, petrophysical variables, and so on (Janiga et al., 2019). Hence, to overcome the problems and gain maximum economic profits, the optimization of well placement is required (Rahim and Li, 2015). In well placement optimization problems, reservoir flow simulation is normally used to integrate geological (static) and dynamic data, and evaluate the objective functions which are normally related to the economic performance of the field (i.e., the NPV). However, reservoir uncertainties strongly affect the accuracy and reliability of reservoir simulation and optimization outcomes. In the following subsections, we explained the required concepts in the robust well-placement optimization under uncertainties. Reservoir uncertainties arise when there are some constraints in the understanding of the reservoir properties (Hutahaean et al., 2019). Hence, instead of optimizing a deterministic model, robust well placement optimization is performed to optimize the objective functions over a reservoir model set (Badru and Kabir, 2003; van Essen et al., 2009; Yang et al., 2011and Chang et al., 2015). During robust optimization, the decision-maker looks for an optimal risk-weighted solution that has good performance for all reservoir models under reservoir uncertainty (Yang et al., 2011). Reservoir uncertainties can be divided into two groups including (1) geological (static) uncertainties related to geological and petrophysical properties, (2) dynamic uncertainties associated with flow properties, production system accessibility, and oil price fluctuation (Santos et al., 2018a). Static uncertainties in well placement optimization are commonly considered by generating numerous geological realizations (static reservoir models) while dynamic uncertainties are taken into account by building multiple simulation models (dynamic reservoir models). Monte Carlo (MC) and Latin Hypercube (LH) sampling methods are standard tools (Santos et al., 2018b) for generating the geological realization. To combine the static uncertainties with dynamic uncertainties and build the simulation model set, Discretized Latin Hypercube Sampling with Geostatistical realizations (DLHG) has been widely used during the last decade (Almeida et al., 2014; Avansi et al., 2015; Bertolini et al.,2015 and Schiozer et al., 2015).
Selecting representative models for ensemble-based production optimization in carbonate reservoirs with intelligent wells and WAG injection
Production optimization under uncertainty is complex and computationally demanding, a particularly challenging process for carbonate reservoirs subject to WAG injection, represented in large ensembles with high simulation runtimes. Search spaces of optimization are often large, where reservoir models are complex and the number of decision variables is high. The computational costs of ensemble-based production optimization can be decreased by reducing the size of the ensemble with representative models (RM). The validity of this method requires that the RM maintain representativeness throughout the optimization process, where the production strategy changes at each evaluation. Many techniques of RM selection use production forecasts of the ensemble for an initial production strategy, which raises questions about the robustness of the RM. This work investigates approaches to ensure the consistency of RM in ensemble-based long-term optimization. We use a metaheuristic optimization algorithm that finds sets of RM that represent the ensemble in the probability distribution of uncertain attributes and the variability of production, injection, and economic indicators (Meira et al., 2020). Our case study is a benchmark light-oil fractured carbonate with features of Brazilian pre-salt reservoirs and many reservoir and operational uncertainties. We obtained production, injection and economic indicators using different approaches to provide valuable insight for RM selection. We inferred about RM fitness for production optimization based on their adequacy for uncertainty quantification for varying production strategies. Despite the effects of changing decision variables on RM representativity, our results suggest the possible use of RM for ensemble-based production optimizations with limitations related to the estimation of the probabilistic objective function due to mismatches in the probabilities of occurrence. Using production indicators obtained from a base production strategy decreased RM representativeness when compared to RM selection based on a more robust evaluation of reservoir performance using a wide-covering well pattern and no restrictions from production facilities. Finally, our results suggest valid RM selection using production forecasts for intermediate dates of the simulation period, an important contribution for ensembles with very high simulation runtimes. We also provide a broad theoretical background on the uncertain reservoir system and on approaches to obtain reduced ensembles and their
applications.