Abstract: Aiming at the influence of mobile nodes on network topological stability, an adaptive distributed reinforcement learning algorithm is proposed to predict the stable connection of adjacent nodes. Each node uses the method of combining reinforcement learning with adaptive division of learning intervals, uses the received signal strength information between adjacent nodes to determine the connection state between adjacent nodes, and finally predicts the set of neighbor nodes that can maintain stable connection. The simulation results of random walk model under various conditions show that the prediction accuracy is about 95%, which verifies the effectiveness and stability of the algorithm.