To address the challenges of security

stability

and renewable energy integration caused by high-penetration distributed generators in active distribution networks

this study proposes a multi-time-scale collaborative optimization method for active distribution network dynamic reconfiguration and reactive power compensation based on graph reinforcement learning. The method aims to overcome the limitations of traditional approaches in handling high-dimensional mixed- integer nonlinear stochastic optimization problems

such as high parameter sensitivity

low computational efficiency

and insufficient utilization of grid topological information. A two-stage collaborative optimization model is constructed

integrating a stepwise renewable energy curtailment strategy to achieve fine-grained control of both slow- and fast-time-scale devices. Simulation results on a modified IEEE 33-node system demonstrate that the proposed method improves the reward function value by 78.17% compared to the original network

while reducing network losses and voltage deviations by 56.24% and 44.48%

respectively. The findings indicate that the method enables deep coupling of dynamic reconfiguration and reactive power compensation without requiring precise grid parameters

leveraging graph-structured features and multi-time-scale coordination mechanisms. This provides a theoretically innovative and practically viable solution for high-renewable-penetration scenarios

effectively balancing operational efficiency

voltage stability

and renewable energy utilization.