The large-scale of degrees of freedom of the train-track-soil coupling system 3D finite element model leads to low efficiency of dynamic analysis, when the random excitation induced by track irregularity is considered concurrently, the analysis of large samples limits the application of 3D model in train-induced environmental vibration. Aiming at the issue of low efficiency in the random analysis of train-track-soil coupled system due to the random characteristics of track irregularity, the pseudo-excitation method(PEM) is used to reduce the cost of computing of large samples; according to the sparsity of equivalent stiffness matrix of train-track-soil coupling system, large-scale sparse matrix was stored in Compressed Sparse Row (CSR) forma, the symmetric positive definite equation of equivalent static equilibrium is solved with the preconditioned conjugate gradient (PCG) method, the parallel program is developed with the Matlab-CUDA (Compute Unified Device Architecture) hybrid platform. The numerical example shows that the efficiency of solving equivalent static equilibrium equation based on Matlab-CUDA hybrid platform is 86.13 times higher than the multi-point synchronization algorithm, and the total calculation time of the random vibration analysis is greatly reduced, the memory usage is small and it is easy to implement on personal computer; when using PCG method to solve the large sparse linear equations of vehicle-track-foundation soil coupling system, it is suggested to take the acceleration as the control index of convergence accuracy of iteration; the balance between calculation accuracy and efficiency can be achieved by selecting appropriate convergence accuracy of iteration.