To evaluate and analyze the characteristics of fluids within hydrocarbon reservoirs, it is critical to collect representative fluid samples. For this purpose, the application of formation testing and fluid sampling with downhole fluid analysis is now well established in our industry. In this paper, we introduce a new approach and methodology for the real-time analysis and visualization of crucial fluid properties which are measured while conducting formation testing and fluid sampling operations.
The proposed method combines machine learning techniques with artificial intelligence. To begin, the prediction model is trained using a comprehensive global hydrocarbon fluids spectral database, which includes approximately five hundred fluid spectra from a diverse set of hydrocarbon samples. The training method involves performing singular-value decomposition (SVD) on the spectral data within the database. A model is then constructed by retaining the first and second eigenvectors (also known as principal components) derived from the SVD. Once the model is established, the prediction phase involves projecting the real-time spectrometer measurements onto the first and second eigenvectors. By plotting the projection of the first eigenvector against the projection of the second eigenvector, we can determine the state, fluid type and the dynamic evolution of fluid properties within the formation tester flowlines during sampling and testing operations.
The application of this new method offers a range of real-time applications for formation testing and sampling. The method allows for the discrimination between hydrocarbon and non-hydrocarbon fluids, such as water, mud, and contaminants, as well as the identification of pure hydrocarbon fluids within oil-water mixtures. Moreover, by cross plotting the projection data, it establishes a clear separation among various types of hydrocarbon fluids, such as black oil, volatile oil, gas condensate, wet gas, and dry gas, forming the foundation for real-time fluid classification. These distinctions also facilitate the identification of oil-based mud (OBM) filtrate based on distinct projection angles. Additionally, the evolving data projections over time closely correlate with the Gas-Oil Ratio (GOR) of hydrocarbon fluids in the flowline. Furthermore, within a dual-flowline formation tester this method also allows for comparison and determination of which flowline contains cleaner fluid; and provides a means to monitor the progress of sample cleanup with indicators to help control and optimize the sampling process.
In this paper, we provide concrete examples using field date to demonstrate the practical application of the method. These field examples will also serve as invaluable case studies, showcasing the tangible advantages and outcomes achievable through the application of the proposed methodology.