Objectives To address the problems of solution divergence and severe parameter coupling in global navigation satellite system (GNSS) dynamic positioning under confined environments during the construction of super-tall buildings, a deformation extraction model based on topology-constrained decoupling of large height-difference array GNSS is proposed.

Methods In the observation domain, a robust observation model integrating predicted innovation residuals, satellite elevation angles, and signal-to-noise ratio was constructed. In the state domain, a shared residual tropospheric delay parameter was modeled for the GNSS array, and the prior structural physical information was incorporated into the filter measurement update process through an augmented matrix constraint, thereby achieving parameter decoupling.

Results The results show that the multi⁃dimensional feature-weighting model can adaptively enlarge the observation noise standard deviation of distorted signals at high elevation angles by 1.8⁃2.0 times. Under the four-antenna array collaborative scheme, the standard deviation of the residual tropospheric delay decreases to 0.179 mm/s, which represents a reduction of 44.06% compared with the single-node solution, while the ambiguity fixing rate increases by 14.02%. In addition, the 95% circular error probable of three-dimensional deformation is reduced by 49.88%, and the maximum vertical error decreases by 37.78% compared with the linear-constraint solution.

Conclusions The proposed method provides a stable algorithmic framework for long-term dynamic safety monitoring of large-scale complex engineering structures in confined environments.