Abstract: With the increasing number of high-rise buildings, the fault prediction of low-voltage distribution cabinets, as a key component of the power system, has gained increasing attention. Traditional fault diagnosis methods based on manual inspections suffer from slow response times and insufficient accuracy. This paper conducts an in-depth study on fault prediction methods for low-voltage distribution cabinets in high-rise buildings using Bayesian deep learning. By proposing a fault prediction method based on Bayesian deep learning, this approach combines the strengths of Bayesian inference and deep learning to effectively handle data uncertainty and noise. Through a posterior update mechanism, the model dynamically updates and provides the probability of fault occurrence, thus improving prediction accuracy. Case experiments show that this method demonstrates significant advantages in high-rise building power systems, enabling early warning and reducing the risks associated with equipment failure.