Sylvester tensor equation naturally comes from PDE literature. To overcome the curse of dimensionality, we use recently proposed tensor train format for multi-dimensional arrays (tensors) to make the storage and computation cost feasible. By exploiting the structure of Sylvester tensor equation, we develop a random splitting method with Douglas-Rachford iteration for solving it. As a result, our method takes full advantage of the tensor train structure. The numerical experiments are also represented to show the efficiency of proposed method.