A Structured Pruning based MPC for Virtual Coupling Trains via Utilizing WHT-SINDy

Model Predictive Control is an effective approach for high-precision cooperative control of Virtual Coupling Trains. However, complex nonlinear terms in high-fidelity dynamic models lead to a surge in online computational overhead, severely limiting real-time applications in millisecond-level sampling scenarios. To address this accuracy-real-time trade-off, this paper proposes a structured modeling method based on Weak-form High-order Term Sparse Identification of Nonlinear Dynamics, constructing a high-precision yet low-complexity nonlinear realtime model to support MPC optimization. The method first discretizes the continuous train dynamics using the fourth-order Runge-Kutta method to ensure alignment with real-world sampling data (0.1-0.5 s). Subsequently, high-order terms generated via Taylor expansion are treated as candidate operators; weakform sparse regression quantitatively evaluates their contributions to achieve node-level selection and structural-level elimination of redundant terms. Compared with full Taylor expansion and Multidimensional Taylor Network weight sparsification, this data-driven structural optimization strategy significantly reduces computational load and storage requirements while retaining theoretical rigor and approximation capability. Experimental results on the Zhengzhou-Xi'an Railway Line demonstrate that the proposed method successfully achieves the dual goals of low computational latency and low storage consumption while satisfying control accuracy and safety constraints.