<p>Clustering profiles of longitudinal data is a prevalent method employed for analyzing heterogeneity patterns among individuals, aiming to identify clusters based on diverse patterns of the mean progression trajectories. The correlation structure plays a crucial role in longitudinal data analysis, as accurate modeling of this structure enhances the estimation efficiency of mean trajectories. In this paper, we assume that subjects are sampled from a Gaussian mixture distribution, and incorporate regularized bandable precision matrix structure information for each subgroup. In order to identify the latent group structure, we employ concave penalty functions to estimate the pairwise differences of the model parameters derived from finite mixture models. The model parameters and cluster labels are estimated simultaneously using the Expectation-Maximization (EM) algorithm in conjunction with the Alternating Direction Method of Multipliers (ADMM) algorithm. We establish computational convergence and provide a statistical guarantee through demonstrating the asymptotic rate. Numerical studies and real data analysis show improved clustering results and excellent accuracy performance.</p>