Hybrid Genetic Algorithm With k-Nearest Neighbors for Radial Distribution Network Reconfiguration

Distribution network reconfiguration (DNR) is to control switches to reduce system losses, alleviate overloads, and mitigate voltage fluctuations. DNR is challenging due to nonlinear power flow equations and the inclusion of integer variables representing switch states. Besides, radial distribution networks require a forest structure, imposing a challenging constraint on the switch state variables. Genetic algorithm is one of the commonly used methods to solve the problem. However, to derive radial solutions, it is necessary to either continue iterating until a radial solution emerges or apply slow and complex encoding techniques that compromise the similarity between chromosomes and real topologies. In this study, we introduce a k-nearest neighbors (kNN) structure that replaces the original chromosome with the best-performing radial topology closest to the chromosome based on the L1 norm, creating a population consisting of radial solutions at each iteration. We also propose a graph-based method to find the nearest neighbor. By utilizing the proposed kNN structure, radiality can be guaranteed with simple encoding. The proposed method has been tested on two test feeders including both balanced and unbalanced networks. The results demonstrate that our approach consistently yields superior performance gains, effectively showcasing the advantages of integrating the kNN structure into the genetic algorithm framework.