The objective of power system restoration is to re store the load as soon as possible, while the successful start-up of generator is the precondition of load recovery and the emphasis of power system restoration. In this paper, both the generator start-up sequence and the restoration path are considered. A preference multi-objective model which considers the difference of the objectives' importance is proposed. To get the most preferred alternatives, a non-r-dominance sorting genetic algorithm II (r-NSGA-II) is used to solve the model. The reference point determination method and solution set scale controlling method are proposed to improve the algorithm, which makes the solution set scale controllable and the preference to be better expressed. Simulation results show that the proposed method has practical value when solving the generator start-up problem for network reconfiguration.