The service restoration problem in distribution power system is the problem of handling operations that make it possible to supply power from other lines in response to power-system failures or construction, by switching between the opened and closed states of sectionalizing switches. Considerable research has already been conducted with regard to this issue, for purposes such as the minimization of unrecovered load and power-supply loss, consideration of priority sections, and preparation of switching procedures. According to past reports, studies have been divided between those that determine target systems for final recovery and those that provide procedures for switching from the system subject to power failure to a target system. This paper addresses the issue of determining target systems for final recovery in cases when some sections remain subject to power failure (i.e., sound bank capacity < load capacity).According to the latest literature, it has been reported that unrecovered loads involved in these issues can be improved effectively through efficient use of the genetic algorithm (GA) method, simulated annealing (SA), and Pareto optimization. However, since this involves application of a number of methods, the related algorithms are complex. In addition, since large numbers of generations and large individual sizes are needed to use GA, extensive calculations are required. Furthermore, since this approach uses Pareto optimization, it does not include the study of all sectionalizing switches.In this paper, in order to resolve the issues noted above we first apply GA as an optimization method for investigation purposes, to simplify the algorithms used. We also propose an improved GA that uses new methods for GA crossovers and mutations. Next, to confirm this proposed method's effects in reducing computing time, we consider the efficacy of the proposed method by using smaller numbers of generations and individual sizes than have been used in the past.As a result of applying the proposed method, it is possible to improve the system states when failure occurs shown in Fig. 1 to the recovered system states shown in Fig. 2. It is clear that the proposed method, consisting of GA only, is superior in terms of average fitness values. These results also show values clearly superior to those reported in the latest literature and show a clear ability to reduce computation times.