As modern industrial chains grow increasingly complex and time-sensitive, traditional transportation planning methods face efficiency bottlenecks. To address this, we propose a parallelization method based on Sparse Matrix-Vector Multiplication (SpMV) to accelerate the Transportation Simplex Algorithm (TSA) for large-scale transportation problems. Existing approaches primarily leverage data parallelism but underutilize GPU computational potential. This study transforms sequential algorithms into matrix-based representations, combining SpMV and parallel graph algorithms to enhance TSA performance. Key contributions include mapping performance bottlenecks—potential vector calculation and closed-loop search—onto SpMV operations, and converting branch logic into matrix operations to reduce divergence. Experiments on a \(10000 \times 10000\) dataset demonstrate a 20.8x speedup for the parallelized MODI algorithm and a 28x overall speedup for solving the Transportation Problem, significantly improving efficiency.