A novel technology that combines the benefits of speed of data sciences with the predictivity capabilities of traditional simulation is being applied to model two blocks of a large waterflood project in the Gulf of San Jorge basin in southern Argentina. The tool is being used to provide a prescription of injection water redistribution that optimizes production and reserves development and reduces injection cost. The technology used is called DataPhysics* and combines the robustness of reservoir physics with the speed of data sciences techniques. The process solves a limited number of unknowns in a continuous scale making it several orders of magnitude faster than traditional numeric simulation. The reservoir model is created from raw (uninterpreted) data and is updated continuously allowing for close loop reservoir optimization in real time. Long term predictivity is enabled by the fact that the tool honors the reservoir physics. At the time of writing this paper the recommendations of the predictive model have been implemented in the pilot sector of the field and early positive results have been observed.

Various types of predictive models have been applied over the years to make quantitative decisions for unconventional development plans. These models are either very simple (e.g., type-curves) which ignore the reservoir physics or are too complex (e.g., simulation models) to be able to run for an entire field efficiently. In this paper, we propose a model for design, prediction and optimization of unconventional wells efficiently using a combination of reservoir physics with machine learning methodologies. The proposed model is the amalgamation of the state-of-the-art in machine learning and reservoir physics into a seamless full field model. The physical model ensures that model predictions are always realistic and reliable while the machine learning algorithm allows us to utilize different types of data to make a prediction which cannot be directly integrated into the physical model. The model uses a probabilistic approach to estimate P10-P50-P90 production curves to account for uncertainty in predictions. The data from more than 1800 unconventional wells in a real field is used to train and test our proposed model. The input features are completion design parameters like lateral length, proppant concentration, well spacing, etc., and the output in a full time series of expected oil production from the well. The results show that our modelʼs prediction leads to correlations of more than 0.75 for the test set which is indicative of its good predictive accuracy. The sensitivity analysis of the parameters of the model on the cumulative production shows that volume of injection fluid, length of the lateral and the proppant concentration are among the most important parameters.

As the oil and gas industry embarks on the path to energy transition, pressure from government regulators, investors, and the public in general demand that companies have clear and transparent net-zero goals and that their operational initiatives and plans support such transition efforts. Mature fields present an opportunity to increase production through operational optimization, which at the same time, can also lead to greenhouse gas (GHG) emissions efficiency. This paper presents the application of a novel modeling and optimization technique in a mature waterflood environment. Data Physics is the amalgamation of the state-of-the-art in machine learning and the same underlying physics present in reservoir simulators. These models can be created as efficiently as machine learning models, integrate all kinds of data, and can be evaluated orders of magnitude faster than full scale simulation models, and since they include similar underlying physics as simulators, they have good long term predictive capacity and can even be used to predict performance of new wells without any historical data. The technology was applied to a mature field in the Neuquen basin in Argentina to effectively reduce the amount of water injected into the reservoir with no negative impact on the production. Additionally, a new Carbon Intensity (CI) modeling tool was used to compare the emissions intensity before and after optimization showing a significant improvement in CI achieving three objectives in one single decision: 1) obtain significant water injection reduction with its corresponding impact in injection and water treatment costs; 2) maintaining production compared to the initial decline of the field, improving the top line; and 3) improving the GHG emissions intensity hence the long term benefit to the environment. The paper deals more with the implementation of the technologies than the technologies themselves, assuming that readers unfamiliar with both Data Physics and Carbon Intensity tools will refer to the references section to gain familiarity with these.

Hydrocarbon production optimization is essential in pursuing the best scenarios for economic outcomes. But because of complex and multi-dimensional nature of production processes, thousands of scenarios are possible. Extensive data collection may allow uncovering patterns still unidentified. With on-site computing power increasing, cloud availability, and artificial intelligence evolution, mathematical optimization methods are becoming powerful and accessible. Data type-tailored models are implemented for history matching and prediction of operational efficiency of the asset. This paper presents a comprehensive analysis and comparison of three data type-tailored reservoir modeling methods and their optimization process for waterflooding field cases. The mathematical techniques used were Data-Driven Capacitance Resistance Model (CRM), Numerical Simulators (Data-Physics) coupled to Smart Algorithms Optimizers, and Hybrid Model (Machine Learning Physics-Based). They were compared to 1-identify the benefits of mathematical optimization techniques, 2-illustrate the methods developed to sort out time and computing capacity restrictions, and 3-validate the techniques by comparing the forecast with actual results. The six study cases of different reservoir types in Argentina, Venezuela, and the USA, had different types data availability. Four had no static model. In two cases, field results were available to confirm the accuracy of the forecasted injection and production. The forecasted increase in Net Present Value (NPV) and cumulative oil production (Np) ranged to 30%, and optimized water injection rates decreased by 50%. Traditional modeling techniques yielding unreliable result in one field with hundreds of producing layers and unknown lateral and vertical continuity were solved using a machine learning technique. In some cases, they pointed toward non-intuitive infill drilling sequence and injection water redistribution. Also, they pointed to options that reduce economic risk. The methods yielded many better economic scenarios and increased the flexibility of operationalizing plans. In one field requiring excessive computing power, using time horizons reduction and successive year-by-year optimization yielded 4 times the NPV of the base case. This approach solves objections related to long computing time and system instability. With the three mathematical techniques, the asset value could be continually maximized by a novel implementation of a heuristic decision-making approach that continuously challenge the current scenario. It makes a systematic formulation of conceivable new scenarios, competing through an objective function determining the probity of compared scenarios. The optimization also resulted in an up to 50% decrease in water injection requirements and the same percentual CO2 emissions reduction.

No abstract