With the development of the social economy, the importance of energy and environmental issues is becoming more and more prominent. Distributed power sources (DG) have the advantages of low carbon and environmental protection, flexible control, and low investment cost, but their nodes in the grid are random, which will increase the complexity and uncertainty of grid operation after grid connection. In this paper, we study the planning path of DGs in distribution networks and the protection technologies involved. Firstly, we focus on the impact of DG access on the protection of distribution networks and the impact of different access capacities and access locations of DG on the protection of distribution networks. Second, after studying the current status of the application of traditional genetic algorithms in distribution network fault location, a two-layer planning model based on the modified extreme learning machine algorithm (MPSO-ELM) is proposed, with the load demand guaranteed by the first layer of power allocation and the fault location incorporated into the distribution network protection scheme in the second layer. Finally, simulations are performed to compare and demonstrate the advantages of this MPSO-ELM algorithm. The simulation results show that the MPSOELM algorithm converges faster than the traditional genetic algorithm by 7s, the average operating speed of power nodes is 3.61ms faster than the traditional genetic algorithm by 5.59ms, and the accuracy in fault location is improved by 23.14%. The simulation results verify the comprehensiveness, accuracy, and high efficiency of the ELM algorithm for double-layer planning of distribution networks containing distributed power supplies, which is important for improving the level of power grid construction and avoiding the risk of power grid construction.