Abstract
Reservoir fluid composition is an essential element in the planning and development of petroleum fields. Experimental analysis and computational modeling are performed on reservoir fluid samples to find molar compositions and to understand fluid behavior throughout the life cycle of the reservoir. There are various computational practices to simulate reservoir fluids behavior applying equation of states and other approaches. However, with the aid of the 4th industrial revolution, and with many successful applications of artificial intelligence in the oil and gas industry, an artificial neural network model was built to predict molar compositions for different reservoir fluids.
The expert system described in this work is a tool which can be used to predict molar compositions of reservoir fluids. One of the main challenges in predicting fluid compositions is its complexity and its wide differences. There are huge composition variations between different formations and sometimes within the same formation due to possible existing compartments in the reservoir. Moreover, the compositions vary with several factors such as location, reservoir conditions, and time. Therefore, the inputs to the ANN model should include well location 3D coordinates, reservoir conditions, API gravity, gas specific gravity, average molecular weight, sampling date, as well as reservoir and field information.
The ANN model was trained based on historical experimental data of fluids from different fields and formations. Different cases and error analysis were performed to optimize the prediction accuracy of the developed tool. A feedforward back-propagation algorithm was applied in building the model. Results indicated that a more accurate model could be obtained without grouping any hydrocarbon components. A successful prediction of molar composition was achieved using the built ANN model. This method can add a great value to the reservoir characterization process as the fluid molar composition can now be predicted using simple measurements of fluid properties and other reservoir data that can be easily obtained.
Introducing this ANN model into the process of fluid compositional analysis will optimize the operational process by reducing the time and cost associated with the experimental analysis.
