To solve the premature issue of particle swarm optimization (PSO), this paper introduces the idea of simulated annealing into the dynamic inertia weight adjustment to make the inertia weight decrease with the increase in the number of iterations and avoid the algorithm falling into local optimization. And it uses population fitness variance to judge the particles’ similarity in order to mutate the particle adaptively, i.e. control the mutation number “big then small” and select particles randomly and rebuild displacement. In order to improve the search speed of PSO algorithm, the PSO algorithm and branch exchange algorithm are combined, in which the particle displacement is just the contact switch connecting non-fault power-lossing area and normal area, as for the scheme of forming looped network, the branch exchange algorithm is used to determine the section switch to open. The combination of PSO and branch exchange could ensure the radial operation requirements of distribution grid, and make the network formed by a lower net loss. The example shows that the distribution network fault recovery based on branch exchange-PSO can quickly restore power supply of power failure area.