Optimization of production network systems together with reservoir dynamics is an integrated procedure to maximize hydrocarbon recovery under different facility and field operating constraints. However, existing tools are often unsuitable for continuous model updates and rapid simulations. The objective is to develop a fit-for-purpose integrated asset model by combining a reduced-physics reservoir model called reservoir graph network (RGNet) with a novel surface network optimization workflow for integrated optimization from the reservoir to the separator.
Traditionally, reservoir dynamics for integrated asset modeling is often simplified using a set of tank models, which may not produce accurate forecasts in the presence of strong subsurface heterogeneity, reservoir flooding, varying fluid behavior and other complexities. On the other hand, coupling a full-physics reservoir simulation model into surface network model for continuous optimization is intractable given the expensive calibration and high computational requirements. Using RGNet as the model presenting the reservoir flow balances accuracy and cost to provide optimized production schedule with high confidence. Additionally, we develop a novel non-intrusive proxy-based workflow, referred to as SPN (surface pipeline network), to solve the surface network optimization with high efficiency and robustness. Complex situations can be properly addressed including looped network with indeterminate flow directions. Flow in pipes is described with physical equations that depend on PVT properties, and mixing rules are applied to handle mixing of fluids with different PVT properties.
The integrated optimization is done implicitly in an iterative way by coupling SPN and RGNet at each timestep. For each timestep, RGNet is run to generate IPR curves and VLP proxies for pipes are generated by solving physical equations for pipe-flow modeling under different operational conditions. Network optimization is performed using SPN then. Next, fluid ratios and PVT properties are updated based on mixing rules, and VLP proxies are refined for better accuracy. The process is repeated until the convergence of reservoir and surface facility.
The proposed workflow was tested on a synthetic case and two field cases. The plant inlet separator was chosen as the terminal node with fixed pressure and the choke pressure drop at each well was optimized to maximize production subject to different operating constraints. We tested the SPN performance by solving a large-scale synthetic network and compared the results to a benchmark software. We also tested a complex looped field network with compressors and dynamic SPN flow directions, which cannot be properly handled by the benchmark software. For the field case that coupled SPN and RGNet, the results showed that the entire optimization workflow was efficient with good accuracy and yielded considerable improvement to the objective function, while honoring all boundary conditions and physical constraints.