In view of the common problems such as high noise, severe structural occlusion and local density mutation in the actual collection process of substation point clouds, traditional registration methods are prone to unstable convergence and error amplification in the deviation identification task. To this end, this paper proposes an automatic verification algorithm for component position deviation that integrates Gaussian distribution registration and 3D Gaussian Splatting (3DGS). This method utilizes the Gaussian model to establish the probabilistic geometric representation of the device and constructs a continuous structural field through 3DGS, enabling the registration to maintain a stable optimization direction even in sparse and missing regions. Through the joint parameter update mechanism, the algorithm effectively suppresses the drift phenomenon of the traditional Gaussian model in irregular areas. On typical substation component data, the method proposed in this paper is compared with methods such as ICP, GMM-ICP, and CPD. The experimental results show that the deviation error of the traditional ICP is 17.2 mm, while the method proposed in this paper can reduce the deviation to 3.9 mm. The corresponding RMSE has been reduced from 11.5mm to 2.7mm. In terms of convergence efficiency, the number of iterations has decreased from 42 to 21, making the verification process more efficient. Under complex interference conditions (such as 30% noise and 50% point cloud absence), the mean deviations of the method proposed in this paper are 4.9 mm and 5.4 mm respectively, and the variance always remains around 3.5 mm², which is significantly better than the baseline model. Research shows that the Gaussian+3DGS framework has significant advantages in terms of deviation accuracy, robustness, and convergence stability, and can provide a reliable engineering solution for the automatic verification and intelligent inspection of component position deviations in substations.
