Abstract: To address the imbalance of selectivity and speediness in traditional distribution network protection setting caused by distributed generation penetration and topological dynamic reconfiguration, this paper proposes an adaptive protection setting online tuning strategy combined with meta-reinforcement learning. A three-level architecture is constructed, integrating PMU, SCADA and AMI multi-source measurement information to achieve global state perception, completing data fusion and state modeling based on weighted Kalman filtering, and embedding a meta-reinforcement learning strategy network to realize fast cross-scenario tuning. Taking an active distribution network with distributed generation as the test object, results show the proposed strategy effectively solves the core problems of excessive tuning time and insufficient generalization ability under dynamic scenarios.