Fast history matching and optimization using a novel physics-based data-driven model: An application to a diatomite reservoir with hundreds of wells

History matching and optimization of full-physics models can be computationally expensive since these problems usually require hundreds of simulations or more. For a mature field with many wells and decades of production history, this could be a time-consuming task. In previous studies, a physics-based data-driven network model (GPSNet) was implemented with a commercial simulator that serves as a surrogate and then applied successfully to a diatomite reservoir sector in the San Joaquin Valley (SJV) for rapid history matching and optimization.In this paper, we successfully expand the previous work to a larger sector in the same reservoir with more than three hundred wells with two decades of waterflood history. Additional development for well completions and connection rules were introduced in GPSNet to handle various well types such as work-overs, dual-string injectors, and horizontal wells. Updated GPSNet is also flexible enough to allow local network refinement in order to accommodate special areas of interest in the operation. A flow network construction workflow is established and discussed in the context of a complex waterflood scenario.History matching of the sector generated a model that honors field-level production history and gives reasonable matches on the well-level for both pressure and volumetric data. For optimization, a P50 model is selected to maximize the 5-year Net Present Value (NPV) under operational well/field constraints. Two optimization scenarios were investigated: a. optimize well control for injectors and b. simultaneously optimize injector well control and return-to-production/return-to-injection well counts. Both optimization scenarios have provided an approximate 60% increase in NPV compared to the reference case. This successful application of GPSNet to a large sector with complex waterflood history has demonstrated the flexibility and robustness of GPSNet.