In this paper, the security and early warning methods of digital twin voltage in power system are studied. A Markov model considering power load is constructed, and the risk assessment method of voltage obstacle crossing under uncertain condition is given. Then the source - charge model is established by Monte Carlo method and Gibbs method based on the ultra-high real-time computing performance of the digital twin distribution network. The distribution network is simulated. Examples show that this method can greatly improve the speed of substation potential fault warning, improve the location accuracy, and play a crucial role in substation potential fault troubleshooting. Using Markov chain method, the source-charge relationship can be accurately described. The Gibbs sampling method is suitable for the forecast scenario in the period of extracting data from the approximate sampling point in the Markov chain. It is feasible to use Gibbs sampling to extract the prediction scenario of the analysis period.