Inferential Reservoir Modelling and History Matching Optimization using Different Data-Driven Techniques

One of the major problems associated with history matching is the non-uniqueness of the solutions. A major flaw in this traditional history matching is that it lacks robustness as it shows a bias to the production data being matched while neglecting the mechanics governing other production data and such solutions generated are erroneous and gives a poor representation of the reservoir being matched.In this study, data driven and numerical modeling of a synthetic PUNQS3 reservoir were carried out. Single objective function, aggregated and multi-objective functions were adopted for the reservoir history matching. A proxy model was developed with data generated from a reservoir simulator using Artificial Neural Network (ANN) and the Response Surface Methodology (RSM). Firefly Optimization (FFO), Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) algorithms were used for the history matching process.The results showed that the history matching process was strongly influenced by porosity and permeability. The interaction between the two was also established. The ANN appeared to provide a better match of the simulated data compared with the RSM. Although aggregated method of optimization is less computational expensive, the multi-objective approach provided a superior history matching optimization. The observed misfit values were 0.074, 0.073, and 0.073 for GA, PSO and FFO algorithms respectively for cumulative oil production history matching. Better predictions were obtained using the FFO and PSO compared with GA for single and aggregated objective function optimization. This work can be extended to investigate the performance of FFO and other recent methods using multi-objective approach and the influence of objective function on history matching.