Unmanned vehicles usually locate themselves based on prior map matching when operating autonomously in industrial scenarios. However, the scene change will affect the positioning accuracy of unmanned vehicles. In view of this, we propose a LiDAR robust positioning and map maintenance architecture for changing scenarios, which includes map matching, positioning optimization, and map maintenance modules. A matching algorithm based on change detection is proposed, which reduces the matching error caused by changing scenarios. A factor graph fusion mode based on LiDAR odometer and prior map matching is designed to improve the robustness of the positioning solution. A filtering method of false detection points based on nearest point search is proposed, which improves the accuracy of change point detection. Finally, we establish a changing scene verification environment through simulation and experimentation and compare the performance of matching based on Loam and the proposed algorithm. The results show that the algorithm can effectively suppress the matching error caused by the scene change, the root mean square error of positioning is better than 3 cm in the actual scenario, and the positioning accuracy is improved by 67. 4% compared with the traditional algorithm.