Insights on 4D Seismic History Matching for a Producing Field by Fitness Landscape Analysis

This research aimed to gain a more profound understanding of 4D seismic history matching (SHM) by viewing it from the perspective of the search space, or fitness landscape, of the optimization algorithm. In previous work, new and classical methods for fitness landscape analysis (FLA) were used to characterize the landscapes of synthetic reservoir models. This has been extended to a real SHM problem for a producing field in the North Sea. This was challenging because of the problem's high dimensionality and the computational cost of fitness evaluations. Nevertheless, progress has been made towards characterizing the topology of SHM fitness landscapes for a real producing field. Fitness landscapes for two reservoir-model configurations, two landscape-sampling methods, and six combined objective functions, based on production data and 4D seismic data, were compared. Four FLA methods were used to characterize and visualize the topology of the fitness landscapes. The results indicated that the landscapes are complex, multimodal surfaces, which appear to be ill-posed and have non-unique solutions. This may help to explain the results seen in the industry. The results also confirmed that the topology of the landscapes depends on the model's definition, its optimization parameters, and the objective function.