This paper addresses the challenge of insufficient localization accuracy for autonomous wheel loaders in mixing station environments by designing a robust SLAM system that fuses LiDAR and IMU data. The system comprises IMU integration and point cloud de-skewing in the front-end, local map construction and matching based on ikd-Tree, and global consistency updates through prior map matching and factor graph optimization. The front-end constructs an IMU kinematic model for state propagation and builds a local planar observation model using LiDAR data. In the back-end, a fast matching algorithm based on descriptors is introduced for prior map constraint detection, and high-precision state estimation is achieved through an Iterated Extended Kalman Filter (IEKF). Finally, by jointly optimizing IMU, odometry, and loop closure factors in a factor graph, the system significantly enhances robustness and accuracy in dynamic and weak localization environments. This study provides an effective solution for autonomous localization of construction machinery in complex semi-structured environments.