Multi-view data often suffer from feature space heterogeneity and noise interference, which severely affect the stability and accuracy of subspace clustering. To address the limitations of existing methods in handling view dimension differences and low-quality view impacts, this paper proposes a decoupled projection and shared structural consistency multi-view subspace clustering method (DP-SC-MVC). The model designs independent linear projection matrices for each view to achieve feature decoupling and dimension alignment. It simultaneously learns a shared representation across views and imposes low-rank constraints to maintain structural consistency. A dynamic weighting mechanism is introduced to adaptively adjust the contribution of each view, suppressing the interference from inferior views. Additionally, -norm regularization is incorporated to enhance robustness against noise and outliers. The unified optimization framework is solved using the augmented Lagrange multiplier method and alternating direction minimization. Experimental results on multiple multi-view benchmark datasets demonstrate that the proposed method significantly outperforms existing mainstream multi-view subspace clustering methods in terms of clustering performance. The DP-SC-MVC method effectively addresses the issues of feature space heterogeneity and noise, demonstrating strong practical value and broad application prospects.