History matching with full-fidelity reservoir models could be a computationally expensive task. We have previously implemented a novel physics-based data-driven surrogate model (GPSNet) and used it for rapid history matching and optimization for waterflood and steamflood applications in mature fields. In this work, we construct a GPSNet model with customized network configurations to capture reservoir volume and connectivity distributions with associated uncertainties and demonstrate a successful application to a deep-water giant gas reservoir in Northern Carnarvon Basin offshore Australia.
This GPSNet model incorporates both current wells in-production and pseudo wells placed across the reservoir. A flow network is constructed using a network of one-dimensional (1-D) connections that link production-production wells, production-pseudo wells, and pseudo-pseudo wells. Both pseudo wells and connections in the network were designed following our best understanding of reservoir connectivity and uncertainty. Connections are grouped into different categories to account for regional variations at different scales. Reservoir properties are calibrated from historical data through a sequential history match process typical for GPSNet modeling: The average reservoir properties of each connection group were first inferred using Design of Experiment (DoE) and Generic Algorithm (GA); Then, excellent history matching results were achieved by adjusting pore volumes and transmissibilities to match build-up pressures and by adjusting completion-level producer well index and non-Darcy effects to match flowing bottom hole pressures using Ensemble Smoother with Multiple Data Assimilation (ESMDA).
With 200 connections among 39 producers and pseudo wells, the typical run time of this GPSNet model is several minutes which is almost two orders of magnitude faster than a full-physics 3-D model. The integration with a commercial simulator also makes it feasible to extend GPSNet to depict complex processes such as non-Darcy effects and well control logic. GPSNet is also flexible and can be customized to capture regional variations with different uncertainty levels which was a key factor in improving the history matching quality. The history-matched models can be used for fast and reliable decision-making as well as reservoir management.
