In recent years, Machine Learning (ML) has become a buzzword in the petroleum industry, with numerous applications which guide engineers in better decision making. The most powerful tool that most production development decisions rely on is reservoir simulation with applications in multiple modeling procedures, such as individual simulation runs, history matching and production forecast and optimization. However, all of these applications lead to considerable computational time and computer resource-associated costs, rendering reservoir simulators as not fast and robust enough, and thus introducing the need for more time-efficient and intelligent tools, such as ML models which are able to adapt and provide fast and competent results that mimic the simulator’s performance within an acceptable error margin. In a recent paper, the developed ML applications in a subsurface reservoir simulation were reviewed, focusing on improving the speed and accuracy of individual reservoir simulation runs and history matching. This paper consists of the second part of that study, offering a detailed review of ML-based Production Forecast Optimization (PFO). This review can assist engineers as a complete source for applied ML techniques in reservoir simulation since, with the generation of large-scale data in everyday activities, ML is becoming a necessity for future and more efficient applications.

In recent years, machine learning (ML) has become a buzzword in the petroleum industry with numerous applications that guide engineers toward better decision making. The most powerful tool that most production development decisions rely on is reservoir simulation with applications in numerous modeling procedures, such as individual simulation runs, history matching and production forecast and optimization. However, all these applications lead to considerable computational time- and resource-associated costs, and rendering reservoir simulators is not fast or robust, thus introducing the need for more time-efficient and smart tools like ML models which can adapt and provide fast and competent results that mimic simulators’ performance within an acceptable error margin. The first part of the present study (Part I) offers a detailed review of ML techniques in the petroleum industry, specifically in subsurface reservoir simulation, for cases of individual simulation runs and history matching, whereas ML-based production forecast and optimization applications are presented in Part II. This review can assist engineers as a complete source for applied ML techniques since, with the generation of large-scale data in everyday activities, ML is becoming a necessity for future and more efficient applications.

According to the roadmap toward clean energy, natural gas has been pronounced as the perfect transition fuel. Unlike usual dry gas reservoirs, gas condensates yield liquid which remains trapped in reservoir pores due to high capillarity, leading to the loss of an economically valuable product. To compensate, the gas produced on the surface is stripped from its heavy components and reinjected back to the reservoir as dry gas thus causing revaporization of the trapped condensate. To optimize this gas recycling process compositional reservoir simulation is utilized, which, however, takes very long to complete due to the complexity of the governing differential equations implicated. The calculations determining the prevailing k-values at every grid block and at each time step account for a great part of total CPU time. In this work machine learning (ML) is employed to accelerate thermodynamic calculations by providing the prevailing k-values in a tiny fraction of the time required by conventional methods. Regression tools such as artificial neural networks (ANNs) are trained against k-values that have been obtained beforehand by running sample simulations on small domains. Subsequently, the trained regression tools are embedded in the simulators acting thus as proxy models. The prediction error achieved is shown to be negligible for the needs of a real-world gas condensate reservoir simulation. The CPU time gain is at least one order of magnitude, thus rendering the proposed approach as yet another successful step toward the implementation of ML in the clean energy field.

In recent years, Machine Learning (ML) has become a buzzword in the petroleum industry with numerous applications which guide engineers in better decision-making. The most powerful tool that most production development decisions rely on is reservoir simulation with applications in numerous modeling procedures, such as individual simulation runs, history matching and production forecast and optimization. However, all these applications lead to considerable computational time and computer resources associated costs, rendering reservoir simulators as not fast and robust enough, thus introducing the need for more time-efficient and smart tools, like ML models which are able to adapt and provide fast and competent results that mimic the simulator’s performance within an acceptable error margin. The first part of the present study (Part I) offers a detailed review of ML techniques in the petroleum industry, specifically in subsurface reservoir simulation, for the cases of individual simulation runs and history matching, whereas the ML-based Production Forecast Optimization applications will be presented in Part II. This review can assist engineers as a complete source for applied ML techniques since, with the generation of large-scale data in everyday activities, ML is becoming a necessity for future and more efficient applications.

Flash calculations are essential in reservoir engineering applications, most notably in compositional flow simulation and separation processes, to provide phase distribution factors, known as k-values, at a given pressure and temperature. The calculation output is subsequently used to estimate composition-dependent properties of interest, such as the equilibrium phases’ molar fraction, composition, density, and compressibility. However, when the flash conditions approach criticality, minor inaccuracies in the computed k-values may lead to significant deviation in the dependent properties, which is eventually inherited to the simulator, leading to large errors in the simulation. Although several machine-learning-based regression approaches have emerged to drastically accelerate flash calculations, the criticality issue persists. To address this problem, a novel resampling technique of the ML models’ training data population is proposed, which aims to fine-tune the training dataset distribution and optimally exploit the models’ learning capacity across various flash conditions. The results demonstrate significantly improved accuracy in predicting phase behavior results near criticality, offering valuable contributions not only to the subsurface reservoir engineering industry but also to the broader field of thermodynamics. By understanding and optimizing the model’s training, this research enables more precise predictions and better-informed decision-making processes in domains involving phase separation phenomena. The proposed technique is applicable to every ML-dominated regression problem, where properties dependent on the machine output are of interest rather than the model output itself.

An “energy evolution” is necessary to manifest an environmentally sustainable world while meeting global energy requirements, with natural gas being the most suitable transition fuel. Covering the ever-increasing demand requires exploiting lower value sour gas accumulations, which involves an acid gas treatment issue due to the greenhouse gas nature and toxicity of its constituents. Successful design of the process requires avoiding the formation of acid gas vapor which, in turn, requires time-consuming and complex phase behavior calculations to be repeated over the whole operating range. In this work, we propose classification models from the Machine Learning field, able to rapidly identify the problematic vapor/liquid encounters, as a tool to accelerate phase behavior calculations. To set up this model, a big number of acid gas instances are generated by perturbing pressure, temperature, and acid gas composition and offline solving the stability problem. The generated data are introduced to various classification models, selected based on their ability to provide rapid answers when trained. Results show that by integrating the resulting trained model into the gas reinjection process simulator, the simulation process is substantially accelerated, indicating that the proposed methodology can be readily utilized in all kinds of acid gas flow simulations.