In order to improve the speed and fault tolerance of the distribution network with distributed generator (DG), this paper uses highly parallel and robust genetic algorithm to solve the over-reliance on the initialized harmony memory of basic harmony algorithm, and improves the data processing way of harmony algorithm. According to the structure characteristics of the distribution network with DG, the "tree" division method is used to decrease the function solving degree. In order to improve the fault tolerance, the minimum set theory is used to improve fault location evaluated function. The optimized harmony search and basic harmonic algorithm are used respectively in the same distribution network fault location model for analysis and comparison. Results show that the optimized harmony search algorithm can effectively improve the convergence speed and precision, and it has good fault-tolerance as the lack or distortion of the fault information.