The security of power grid is closely related to national security and social stability. With the development of smart grid and big data, grid data security has ushered in new challenges. The data provenance is one of the key issues. Considering the complexity of grid data storage types, this paper designs a virtual tuple-based data provenance scheme. This scheme firstly embedding the watermark with user identity through virtual tuple injection. Then, the data with watermark is distributed in the form of table splitting, which makes it possible to quickly locate the suspect user when a data leakage event occurs. Finally, the leaker is identified through watermark extraction and comparison. The simulation experiments show that our scheme has good robustness against insertion, replacement and deletion attacks.