A Multi-sensor Fusion Approach for Susceptibility Analysis in Smart Substation based on Deep Bayesian Network

Abstract

Since the single-sensor scenario for online detection and fault diagnose might raise the risk probability of false positive or negative, a multi-sensor fusion technique is introduced to enhance the performance of susceptibility analysis in smart substation. With the help of recurrent neural network (RNN) and Bayesian neural network (BNN), the susceptibility of smart device and systems can be numerically evaluated with the recorded time series. As the deep Bayesian network transfer probability distribution along the model forward, the posterior probability distribution can be acquired based on the multi-sensors and historical data. A simulated case of very fast transient overvoltage (VFTO) measured by two voltage sensors is performed, demonstrating the applicability of the proposed method.