AI Based Water-in-Oil Emulsions Rheology Model for Value Creation in Deepwater Fields Production Management

This work aims to present a new paradigm in the Exploration & Production (E&P) segment using Artificial Intelligence for rheological mapping of produced fluids and forecasting their properties throughout the production life cycle. The expected gain is to accelerate the process of prioritizing target fields for application of flow improvers and, as a consequence, to generate anticipation of revenue and value creation. Rheological data from laboratory analyses of water-in-oil emulsions from different production fields collected over the years are used in a machine learning framework, which enables a modeling based on supervised learning. The Artificial Intelligence infers the emulsion viscosity as a function of input parameters, such as API gravity, water cut and dehydrated oil viscosity. The modeling of emulsified fluids uses correlations that, in general, do not represent the viscosity emulsion suitably. Currently, an improvement over empirical correlations can be achieved via rheological characterization using tests from onshore laboratories, which have been generating a database for different Petrobras reservoirs over the years. The dataset used in the artificial intelligence framework results in a machine learning model with generalization ability, showing a good match between experimental and calculated data in both training and test datasets. This model is tested with a great deal of oils from different reservoirs, in an extensive range of API gravity, presenting a suitable mean absolute percentage error. In addition to that, the result preserves the expected physical behavior for the emulsion viscosity curve. Consequently, this approach eliminates frequent sampling requirements, which means lower logistical costs and faster actions in the decision making process with respect to flow improvers injection. Moreover, by embedding the AI model into a numerical flow simulation software, the overall flow model can estimate more reliably production curves due to better representation of the rheological fluid characteristics.