With the transition from traditional centralized power systems to distributed energy systems, harmonics are produced due to the penetration of renewable energy, resulting in additional losses. Distribution network loss accounts for more than half of the total system loss and is the focus of network loss reduction. A network reconstruction method based on improved binary particle swarm optimization algorithm was proposed to reduce distribution network loss. Firstly, considering the influence of harmonic wave on network loss and the harmonic effect of the line under high frequency current, the line impedance was modified. Then, the total loss of the network was calculated using the corrected impedance, and a probabilistic reverse learning approach suitable for binary algorithms was creatively proposed, integrating the theory of the good point set to obtain uniform and diverse initial particles. Finally, taking the modified 33 node power system as an example, the optimization objective was to minimize the total loss and obtain the optimal topology structure of the distribution network. The experimental results show that the distribution network reconstruction considering harmonic factors has played a good role in reducing the loss of distribution network.