Speaker
Description
Accurate prediction of reservoir fluid behavior is fundamental for efficient reservoir management and optimized hydrocarbon recovery. Traditional reservoir engineering approaches, while robust, often face challenges in capturing complex fluid dynamics and integrating diverse datasets from well testing and fluid property analysis. This research presents a hybrid artificial intelligence (AI) model that synergizes machine learning techniques with physics-based reservoir engineering principles to predict reservoir fluid behavior and performance with enhanced accuracy.
Using well testing data combined with pressure-volume-temperature (PVT) fluid property measurements, the proposed model employs advanced deep learning architectures constrained by reservoir physics to maintain physical consistency in predictions. The training dataset incorporates key reservoir fluid parameters such as viscosity, saturation, and pressure transient responses, enabling the model to dynamically forecast fluid flow characteristics and production performance metrics.
Validation against historical production data and reservoir simulation results demonstrates that the hybrid AI model not only improves prediction accuracy but also provides a practical framework for near real-time fluid behavior forecasting. The integration of explainable AI techniques allows for model interpretability, bridging the gap between conventional reservoir engineering and modern data-driven approaches.
This study contributes a novel methodology that enhances reservoir fluid characterization, supports decision-making for reservoir development, and offers a pathway for deploying AI-assisted tools in petroleum reservoir engineering.