Traditional distribution network service restoration cannot completely satisfy the requirements of N-1 security verification of transformer in station, and the corresponding mathematical model is constructed. Without considering the preference of decision maker in the iterative process, the classic multi-objective evolutionary algorithm, based on Pareto dominance criteria and crowding distance sorting method, is of slow convergence speed. This paper proposes a multi-objective binary particle swarm optimization based on g dominance and technique for order preference by similarity to an ideal solution (TOPSIS) to solve distribution network service restoration. The reference point of g dominance can design flexibly according to the requirements of service restoration, and TOPSIS makes the preference of decision maker into the iteration. These improved measures are of great significance on improving the quality of solution and enhancing convergence velocity. Finally, an example is illustrated to verify feasibility and validity of the proposed method.