Identifying key Connected and Autonomous Vehicles (CAVs) to influence Human-Driven Vehicles (HVs) offers a robust solution for optimizing mixed traffic efficiency and safety. However, traditional pinning node selection primarily relies on vehicle spatial distribution, usually neglecting the latent impact of CAV behavioral characteristics on platoon dynamics. Furthermore, large-scale Vehicle-Road-Cloud Integrated (VRCI) networks impose stringent requirements on both the number of pinning nodes and real-time decision-making. To address these challenges, this paper proposes a Pinning Node Identification Model based on RatioCut Spectral Clustering (RCSC) for Vehicle-Road-Cloud Integrated tunnel scenarios. This method constructs a multi-dimensional feature vector by integrating coupled platoon structures, trajectory consistency, and safety attributes. By leveraging spectral clustering, the model precisely identifies the optimal pinning control targets. To ensure millisecond-level response in cloud-coordinated decision-making, the RatioCut method is employed to reduce the dimensionality of the Laplacian matrix’s eigenvalue decomposition, significantly accelerating computation. Experimental results show that under identical pinning control algorithms, the RCSC strategy outperforms alternative schemes, achieving performance gains of at least 2.3%, 3.2%, 6.2%, and 3.5% in the standard deviation of velocity difference, acceleration, inter-vehicle spacing, and relative velocity, respectively. These findings underscore the efficacy of RCSC in enhancing the overall operational efficiency and consensus of mixed traffic platoons in complex tunnel environments.