Large-scale distributed photovoltaic grid-connection has brought serious power back-feeding to the medium-voltage distribution network, resulting in node voltage limit violations and reverse overload of distribution transformers in the medium-voltage distribution network. An evaluation model for distributed photovoltaic hosting capacity in medium-voltage distribution networks was established, and an improved simulated annealing-particle swarm optimization (SA-PSO) algorithm was proposed. The system power flow balance, node voltage deviation, reverse load rate of distribution transformers, and line current carrying capacity were taken as constraints, and the distributed photovoltaic equivalent grid-connection capacity was taken as the objective function, which was the difference between the distributed photovoltaic grid-connection capacity and the system network loss. The distributed photovoltaic hosting capacity calculation was performed on the IEEE33 system and an actual medium-voltage distribution network. Results showed that the established distributed photovoltaic hosting capacity evaluation model was suitable for evaluating the stability of node voltage and safe operation of distribution transformers in medium-voltage distribution networks. Compared with other algorithms, the improved SA-PSO algorithm improved the convergence speed and optimization ability of the evaluation model calculation. Under the same constraints, the obtained line distributed photovoltaic hosting capacity was higher and the system network loss was lower, compared with those of other algorithms.