The state of power grid is determined by states of its systems, conditioned by their mutual influence of different nature. These influences cause the change of state of each system during grid life cycle. The proposed approach is based on application of Bayesian Belief Network, where nodes represent different grid systems, and links are stipulated by different types of influences (physical, informational, geographic, etc). The grid safety is evaluated on its criticality. The criticality and influence are treated as the linguistic values. It is suggested to evaluate criticality of system, considering the change of criticalities of all connected systems. Conditional probabilities are also represented by linguistic values. To demonstrate the approach to the grid safety analysis, Russian Sayano-Shushenskaya hydro power station accident is reviewed. Keywords: Power grid, Computing with words, Safety, influence, Bayesian Belief Network
