Autonomous topology planning for distribution network expansion: A learning-based decoupled optimization method

The distribution of urban electricity load is highly uneven in both space and time, posing significant challenges to the problem of distribution network expansion planning (DNEP). An autonomous topology planning method for distribution network expansion is proposed to address that, driven by a learning-based decoupled optimization approach. The planning problem is decoupled from the impact of distribution network operation and used as reinforcement cost prediction for autonomous planning after the generation of topology schemes. First, a reinforcement model of the original distribution network is built that considers the actual distance calculation method of load nodes. Then an effective method for generating historical operation samples is designed, followed by the development of a learning-based reinforcement cost prediction model for the original distribution network based on operation state learning using XGBoost, which enables the prediction of said costs under different scenarios. Utilizing a distribution network topology generating method driven by the adjustable weight minimum spanning tree (AWMST) algorithm, an autonomous topology expansion planning strategy generation method employing the learning-based decoupled optimization approach is established. The feasibility and superior analysis of the proposed method is verified and analyzed on an actual urban distribution network case with simulation scenarios. The results show that the proposed method is faster and more flexible to deal with the distribution network planning problem brought about by rapid load growth in urban distribution networks.