Since most of the current tensor decompositions use iterative algorithms, these algorithms can theoretically converge to the global optimal solution, but in practice, due to noise and other factors these algorithms tend to converge to different local optimal solutions, so there is a need to analyse the reproducibility of tensor decompositions in order to ensure the reliability of the algorithmic results. Based on this, in this paper, according to the previously developed coupled tensor decomposition algorithm based on federated learning (FCNCP), its decomposition results in simulated and real data are subjected to reproducibility analysis, so as to verify the stability of the FCNCP algorithm. In this study, we have used both simulated and real data for our experiments. The reproducibility metrics of the 50 decomposition results of the simulated data reached 1.0. Whereas the results of 50 decompositions using the fifth-order ERP tensor data collected by applying proprioceptive stimuli to the right and left hands, the reproducibility metrics for each of the 10 components of interest selected were 0.92, 0.8, 0.82, 0.8, 0.82, 0.57, 0.94, 0.91, 0.79, 0.94 respectively. This shows that our previously studied FCNCP algorithm is stable and effective.