Accurate and fast calculation of the effective elastoplastic properties of lattice materials is a key prerequisite for achieving reliable homotopy mapping of microstructure-macro mechanical response and supporting cross-scale parameter inversion, structural optimization and engineering safety margin assessment[1,2]. However, it becomes prohibitively expensive for large truss lattices with slender members, where refined solid-element DNS suffers from exploding degrees of freedom and iterative plasticity costs[3]. We present a data-driven reduced-order framework—FEM-Cluster based Analysis (FCA)[4]—specialized for truss lattices that blends cluster-based model reduction with physics-informed simplifications to deliver fast, accurate elastoplastic homogenization. In the offline phase, truss elements are clustered by axial strain signatures and orientation correlations, and a self-equilibrated cluster interaction matrix (CIM) is assembled using global-frame transformation to preserve stress compatibility. In the online phase, we formulate an energy functional under the Principle of Cluster Minimum Complementary Energy (PCMCE)[5] and employ a return mapping algorithm for incremental elastoplastic solutions. Local compressive buckling of slender struts is checked, and a rapid procedure is devised to construct the initial yield surface of truss lattices. Compared with DNS, the proposed truss-FCA markedly reduces computational cost while retaining high accuracy in predicting effective elastoplastic properties. By leveraging PCMCE and cluster-wise thermal-eigenstrain loading for CIM construction, the framework achieves substantial speedups across diverse lattice topologies and infill configurations without sacrificing predictive fidelity.