Service restoration of distribution systems is a major controlling method of realizing self-healing for smart distribution. This paper presents an improved genetic algorithm for distribution network service restoration reconfiguration. An encoding strategy based on loop is proposed according to the structure character of distribution network with distributed generation, so the invalid solutions that represent rings or islands can be reduced. The adaptation crossover rate and mutation rate make the algorithm alterable based on population evolution, which protects the optimized individual. The adaptive population size can improve the convergence properties of the genetic algorithm, based on the individual lifetime and size control. The proposed algorithm is simulated in the IEEE 33-bus system, and it shows the access of distributed generation can effectively reduce the network loss. Compared with the other algorithm, the results show that it is more stable and has faster convergence.