Urban rail systems often struggle with over-saturated passenger flows during peak hours, where even operating trains at the minimum headway remains insufficient. The critical bottleneck that limits further improvements in transportation capacity lies in the platform length, which constrains the maximum train composition length. To address this, this paper investigates the strategic use of Extra-Long Train Compositions (ELT) with Flexible Alignment (FA), which allows trains to protrude beyond platforms. We establish a Mixed-Integer Linear Programming (MILP) model to jointly determine train compositions length and train-platform alignment relationships. Due to the complexity of the model, it’s hard to be solved by commercial solvers. Therefore, we adopt the Simulated Annealing (SA) algorithm to explore the vast solution space and evaluate the objective value of each candidate solution. This approach is applied to a scenario case based on the 7:00–11:00 morning peak for comparative experiments. The computational results demonstrate that, compared to traditional scenarios, the proposed ELT strategy can significantly reduce passenger waiting time and the SA algorithm can effectively handle large-scale problems.