Comparative Study of a Fast History Matching and Optimization Tool with Sector Model and Full Field Model

Abstract Field management often faces the challenge of full-physics models in history matching and optimization being too resource-intensive, as they demand numerous simulations. To overcome this, we introduced the General Purpose Simulator powered Network model (GPSNet), a new data-driven flow-network model, for fast history matching and optimization. Previously, GPSNet was applied to construct a full field model with over 1,000 wells (Ren et al., 2024). This paper presents a revised network model tailored for a sector model in the same field, along with a comparative analysis of both models. GPSNet constructs a flow network using 1-D connections between well completions to simulate multi-phase flow through the reservoir and wellbore. The network is simulated with a commercial simulator where it will be mapped onto an equivalent Cartesian model and the grid properties are calibrated to historical production data. An iterative history matching process with Ensemble Smoother with Multiple Data Assimilation (ESMDA) fine-tunes the model to align with production data. Finally, the optimal match is selected for well control optimization to maximize oil production rates within the operational limits of the field. The approach was tested on a waterflood optimization for a sector model with 45 wells. The network model's history matching aligns with the field-level production history and accurately reflects most individual wells. A comparison reveals that producer well matches are significantly improved in the sector model over the full field model. The best scenario for well control optimization is achieved through both well water injection rates and liquid production rate optimization, highlighting the benefits of integrating producer control beyond mere injection rate adjustments. The new network model accelerates the process by using only well volumetric production/injection data and its approximate trajectory, allowing for quick construction and updates. It operates significantly faster than traditional full-physics models, taking mere seconds to run due to fewer grid blocks. Additionally, we demonstrate the necessity and added advantage of incorporating producer controls into well control optimization instead of solely injection rate reallocation.