This paper presents a method for multi-objective distribution network reconfiguration. By introducing the gardener diagram to analyze network structure, the switch combinations that deteriorate objectives are eliminated, and the search space is reduced. Taking power loss, minimum node voltage, number of operation switches, and the load rate dispersion as objectives, the concept of the Pareto optimal solution is utilized to balance the conflicts among multiple objectives. By learning existing information to generate adaptive parameters, an improved harmony search algorithm is proposed to solve this nonconvex model. Case studies on the 33-node system validate the efficiency of the proposed method.