Under the large amount of distributed energy access, the operation state of the distribution system becomes more and more complex, and the traditional state sensing method is difficult to meet the demand for high precision. This paper aims to realize real-time state sensing and early warning of the new distribution system, calculates the equivalent electrical distance of nodes, and adopts the voltage-active power sensitivity to characterize the electrical distance between nodes, so as to closely connect the nodes in the region. On this basis, a Bayesian network is introduced to design a time series simulation method for distribution network components and system state respectively, so as to perceive the distribution network system state in real time. Integrating multiple sources of distribution information, including grid monitoring, user feedback and environmental detection, and constructing a fault indicator fusion information matrix, real-time monitoring and fault warning of the distribution network system state are realized. In the practical application of this method, the abnormal situation can be sensed in 2s and the warning response can be activated in 3s, which has high warning speed and response sensitivity.