With a large number of distributed generation connected to the distribution network, on the one hand, the network loss of the distribution network increases; on the other hand, the fluctuation of its output makes it difficult to absorb the distributed generation. Therefore, a robust optimization method of distributed generation output based on second-order sensitivity network loss is proposed in this paper. Firstly, the calculation formulas of active power loss and reactive power loss based on second-order sensitivity are derived by Taylor expansion, which are embedded into the distributed generation output optimization model, and the optimization model of minimizing comprehensive network loss and maximizing distributed generation output is established; Then, the polyhedron uncertainty set is used to describe the distributed generation output, and the robust optimization model of distributed generation output is constructed. The robust optimization model is transformed into a deterministic quadratic programming model through dual transformation. Finally, the simulation results of 70 buses distribution system show that the error of the proposed second-order sensitivity network loss calculation method is small, and the maximum error is only 5%. The robust optimization method can control the conservatism through polyhedron dimension, and the optimization efficiency is high.