Traditional distribution network reconfiguration(DNR) algorithms often suffer from limited real-time perfor-mance and weak generalization due to their reliance on precise physical models.To overcome these limitations

this paper proposes a dynamic reconfiguration strategy for distribution networks based on a dueling deep Q-learning network(dueling DQN).Built upon a deep reinforcement learning(DRL) framework

the proposed method reduces dependency on accurate system parameters and detailed topological information

thereby enhancing its applicability across various distribution network scenarios.A double Q-learning mechanism is incorporated to address the overestimation bias inhe-rent in traditional DQN

effectively avoiding convergence to local optima.Furthermore

to mitigate the training complexity induced by the high-dimensional action spaces

a convolutional neural network(CNN)-based static reconfiguration model is employed to predict the system state and filter the action space accordingly

to address the challenge of high-dimensional action spaces

CNN-based static reconfiguration model is employed to predict the system state and filter the action space accordingly

reducing the computational complexity of learning.Simulation results based on the IEEE 33-bus test system demonstrate that the proposed approach effectively mitigates voltage violations

lowers operational costs

and improves decision-making speed for dynamic reconfiguration.