The invention relates to a method for realizing distribution network reconfiguration based on multiple agents. The method comprises the following steps: (1) network area division is carried out, and agent structure information is initialized; (2) speed and position information of all agents in each area is initialized; (3) power flow and network loss are calculated; (4) whether or not to succeed in a competition and cooperation mechanism of a multi-agent system grid structure is judged; (5) self-learning operation is carried out on the successful agents, and the unsuccessful agents are corrected by a binary particle swarm algorithm; (7) the fitness values of all the agents are calculated; (7) whether the number of iterations is met is judged, the method goes to the next step if the number of iterations is met, or the method returns to step (3); (8) reconfiguration results in the areas are output to a distribution sub-station agent; and (8) the optimal solution of whole-network reconfiguration is output. By adopting the method of the invention, rapid convergence can be realized, and an accurate global optimal solution can be obtained.