Marked by complex physics in fluid flow through fractured porous media, reservoir modeling for unconventional reservoirs is often deemed a challenging task for production forecast and optimization at field scale. Traditional methods such as decline curve analysis (DCA), rate transient analysis (RTA) and numerical simulation are often not adequate to simultaneously meet the requirements of high model fidelity, sustainability, and ability to handle varying field conditions. It is desirable to have a more tractable approach that can be used to quickly model, history match and predict performance of unconventional wells at field scale using routinely available data.
We propose to use a reduced-physics approach called Reservoir Graph Network (RGNet) model for modeling unconventional reservoirs. Inspired by the concept of the diffusive time of flight, RGNet parameterizes a 3D reservoir by a set of 1D grid blocks with interconnecting wells, which enables fast model runs by reducing the system complexity. The set-up of an RGNet model does not need interpretive forward modeling with geological and geo-mechanical characterization of the reservoir. Instead, the model parameters of RGNet are decided by history matching the observed data from the field. The parameters considered in RGNet include relative permeabilities, initial saturations, well indices, cell pore-volumes and transmissibilities. RGNet presents a general framework, where any fluid and rock physics can be incorporated.
We apply the proposed approach to two field cases, one with a single well and the other with three wells with interference. Multiple history-matched models are generated for reservoir characterization, uncertainty analysis, and future predictions. For the examples presented, it takes less than one second to finish one RGNet forward-run. Compared to a full-physics model, an RGNet model is several orders of magnitude faster without losing out on model fidelity.
The proposed workflow overcomes the challenges for reservoir modeling of unconventional reservoirs based on traditional DCA and other physics-based approaches. As RGNet describes a reservoir with general reservoir parameters, the obtained results are physically interpretable and explainable. The advantage of low computational cost makes RGNet suited for fast-paced reservoir management of unconventional reservoirs.