The uncertainty and correlation of distributed renewable energy generators caused by fluctuation has a large impact on power distribution system dispatch. To study the impact on the dispatch strategy of the distribution system considering maximum power supply capacity, this paper builds an uncertainty set of renewable energy output using a data-driven technique. A bi-level robust optimization dispatch model is established considering the worst scenario of renewable energy output. The lower model aims to minimize the generating cost and thus improves the economy of the dispatch strategy. The upper model takes the worst scenario of the renewable energy output into account and improves the robustness of the dispatch strategy. The proposed bi-level model is simplified into a single model by KKT conditions. Then, the single model considers the uncertainties of renewable energy and the maximum power supply capacity and uses a repeat power flow algorithm. The simulation results demonstrate that the proposed method is effective and economical, and helps to improve the maximum power supply capacity of the distribution system. This work is supported by the Science and Technology of China Southern Power Grid (No. ZBKJXM20180068).