An intelligent heuristic-clustering algorithm to determine the most probable reservoir model from pressure–time series in underground reservoirs.

Precise characterization of underground reservoirs requires accurate calculations of the reservoir's petrophysical data and accurate selection of the mathematical model governing the reservoir's dynamic. In this study, we develop a novel heuristic-clustering algorithm, namely GA–DBSCAN–KMEANS, that can be applied over pressure transient data to assess the true reservoir model out of a pool of candidates. In this algorithm, each specific reservoir model is considered a subpopulation in the GA (genetic algorithm). Then, the simultaneous optimization of all the reservoir models is sought using the proposed hybrid algorithm. During the optimization process, the population size of different models will be either decreased, increased, or unchanged based on the average quality match obtained for each model. A combined DBSCAN (density-based spatial clustering of applications with noise)–KMEANS clustering scheme is used to increase the population size for the best reservoir model in each iteration of the GA. The accuracy of the proposed algorithm was verified using several synthetic data and a real field case obtained from the open literature. The tested data were collected from different types of reservoir models, including homogeneous reservoirs, matrix-fracture dual-porosity reservoirs, and fault-limited reservoirs. For uncertainty analyses and to test the performance of the algorithm under large numbers of initializations, Monte Carlo simulations were conducted. Results of the Monte Carlo simulations unveiled high values of P10, P50, and P90 for the probability of the true reservoir model and low values of these statistics for the false reservoir models. This shows that the outcome of the proposed algorithm is not affected by the initial randomization of the solution subspaces; hence, the developed algorithm is a reliable tool in determining the most probable reservoir model from transient well testing data. [ABSTRACT FROM AUTHOR]