Integration of PLT and Tracer Data Using Pattern Recognition for Efficient Assisted History Matching of Heterogeneous North Kuwait Carbonate Reservoir

Abstract

Abstract This paper describes an efficient assisted history-matching (AHM) workflow that integrates production logging tool (PLT) data, streamline trajectories, and tracer data for fields with high-permeability streaks (thief zones). A field case study from North Kuwait of Sabriyah Mauddud (SAMA), a giant carbonate reservoir with more than 400 producers, is presented to demonstrate the application of the new AHM algorithm. In this field, the presence of thief zones was identified during a waterflooding period when water breakthrough occurred much earlier than expected. Data from the PLT and limited core plugs also supported the presence of thief zones in several layers of the reservoir and confirmed the majority of the water was flowing through these thief zones. Therefore, PLT data-derived thief-zone logs were used to populate the distribution of thief zones in the geomodel. However, reservoir simulation demonstrated that cumulative water production was significantly lower than the observed value. Streamline trajectories demonstrated that water was flowing homogeneously in the reservoir. Therefore, a new history-matching algorithm that integrated PLT data directly into the workflow and modified the thief-zone distribution was proposed. The workflow basically identifies the presence of thief zones at the well locations based on PLT data. Streamline trajectories from simulation and available tracer data were used to better understand the connectivity from water injectors to various producers. This information was then used to modify the reservoir model by altering thief-zone distribution. Multiple models were generated by varying the permeability distribution within the generated thief zones and the thickness of the thief zones. These new permeability models were able to produce water at the field level. First, field-level pressure and production rates were matched by adjusting reservoir properties, such as pore volume, oil API, and fault transmissibility. A Markov-Chain Monte-Carlo (MCMC)-based algorithm was used to match well-by-well production rates for oil, water, and bottomhole pressure (BHP). Two-dimensional (2D) discrete cosine transformation (DCT) was applied to the permeability layers to the DCT coefficient domain for optimization purposes. Only low-frequency DCT coefficients that corresponded to thief zones were efficiently sampled during MCMC-based optimization to converge toward an accurate distribution of thief-zone permeability to history match the production data. Therefore, a reduction in sampling space along with the improvement of connectivity helped accelerate convergence. Given the model size and complexity, the optimization converges fairly quickly. Most of the wells demonstrated an excellent match between simulation results and production data of oil, water, and BHP of more than 200 wells with significant production history. PLT data were also closely matched with the simulation production profile at the wellbore. Streamlines of water for the history-matched model demonstrated the water had been flowing through the thief zones. The history-matched model can be further used for better reservoir management and waterflood optimization to improve oil recovery.