Simultaneous Localization and Mapping (SLAM) technology plays a pivotal role in emerging fields such as mobile robots, autonomous driving, and intelligent monitoring and inspection of offshore wind power. The integration of visual/inertial/LiDAR (Light Detection and Ranging) approaches capitalizes on the spatial structure as well as the texture information within the environment, aiming to achieve more robust pose estimation outcomes. Nevertheless, within large-scale scenarios, the issue of error accumulation persists. To further enhance the localization and mapping performance of SLAM systems in complex real-world environments, this paper proposes a graph optimization-based tightly coupled GNSS (Global Navigation Satellite System) RTK (Real-time kinematic) /LiDAR/vision/IMU SLAM approach, leveraging the complementary characteristics among visual/inertial, LiDAR/inertial, and GNSS RTK systems. This method introduces LiDAR depth information into the visual/inertial system and realizes nonlinear optimization of GNSS corner positions, IMU measurements, visual features, and LiDAR point cloud features within a sliding window. By incorporating loop closure detection algorithms (Scan Context and DBoW2), the global localization and mapping accuracy of the SLAM system are further improved. Experimental results on both M2DGR and Real-World datasets demonstrate that this method significantly outperforms mainstream LiDAR-only SLAM, LiDAR/inertial SLAM, and visual/inertial SLAM algorithms in terms of localization accuracy. With the inclusion of loop closure optimization, a globally consistent point cloud map is constructed.