In the construction of a lunar base, autonomous exploration is a challenging task, especially in highly convoluted environments on the lunar surface due to the inability of robots to completely cover the entire exploration environment. To solve this problem, this paper presents a novel autonomous exploration method, including local exploration phase and global exploration phase. In the first phase, map boundary is extended, local frontiers are connected with dynamic expanded Rapidly-exploring Random Tree (RRT) to guide the robot to conduct local exploration. In the second phase, robot is moved to different subareas in the environment, we store the path of the robot to reach each frontier, and constantly updating the path with the movement of the robot, the unexplored frontiers can be reached according to these real-time updated paths. These two phases both dynamically expand over replanning steps. In addition, the LiDAR point cloud two-dimensional grid map processing technology is used at the beginning to reduce the computational complexity. Simulation results demonstrate that this method achieves complete coverage of the environment, and the exploration efficiency is better than that of Motion-primitive-Based exploration Planner (MBP) and Graph-Based exploration Planner (GBP) algorithms.