Hybrid power systems for off-grid sites are commonly designed using simulation. Operating rules for the controller dispatch strategy are defined, and a simulation uses site-specific data to find the minimum-cost system composition that satisfies the site's electricity demand. The resulting design is strongly influenced by the a priori choice of strategy. We present a new approach to design isolated sites: off-grid electricity supply optimization (OGESO). In our approach, the dispatch rules are not fixed, and we simultaneously optimize the composition and the controller strategy. A mixed integer programming model finds the optimal system composition and the optimal dispatch based on one year of deterministic hourly data. Because the optimal dispatch is complex, it is analyzed using data mining to find a practical controller strategy. Computational results on diesel-wind-battery hybrid systems demonstrate the benefits of the proposed approach.