Reservoir management based on unstructured grid reservoir simulation is improved with machine learning based intelligent automation. Reservoir heterogeneity, geological internal boundary features and well geometry complexity are taken into account to automatically detect well zone and focusing reservoir area by calculating the region-of-interests in the model and defining cell spacing for grid coarsening and refinement in the reservoir. Data points for wells in the reservoir are grouped into reservoir regions according to datasets organized as a convex hull, which is a minimum convex set in spatial geometry which encloses the totality of such data points. The reservoir regions form a basis for grid spacing utilized in the reservoir model.