The structure of distribution network is complex, including point-like distribution network scheme and line-like distribution network scheme, and the network structure of different regions and different load characteristics is very different, so it is difficult to achieve reliable power supply. Therefore, the optimization method of power supply scheme of distribution network based on genetic algorithm improved traversal recurrent neural network is studied. The partial density of the peak load in the distribution network and the nearest length to the high-density point are calculated, and a load peak clustering model is constructed; with the objective function of maximizing the net profit asset value of the distribution network and maximizing the power supply radius after optimizing the peak load of the distribution network, a multi-scale power supply scheme for the distribution network is constructed; the genetic algorithm is used to solve the load peak clustering model and the multi-scale power supply scheme of the distribution network, and the preliminary optimization results of the multi-scale power supply scheme of the distribution network is obtained; the initial optimization results of the power supply scheme is innovatively used as the input for traversing the recursive neural network, and the optimal multi-scale power supply scheme is output for the distribution network. The experiment shows that the proposed method can effectively cluster the peak load of the distribution network, with a power supply radius of 3.007 km, a calculated total cost of 200 600 yuan, and a substation capacity to load ratio of 2.01. After he power supply scheme score is above 9.0, and the total cost of the optimal scheme is only 2 million yuan. It obtains a multi-scale power supply scheme for the distribution network that takes into account both power supply radius and economy.