An approach to investigating the states of complex industrial networks with adaptive topology using network motifs - statistically significant subgraphs of a larger graph - is proposed. The analysis presented addresses the ability of network motifs to characterize system performance and the possibility of their application to short-, medium-, and long-term prediction of system states. Using the Smart Grid network structure as an example, a directed graph is modeled, in which the most common motifs are searched, several attack scenarios on network nodes are simulated and a network state prediction is built. The results of experimental studies confirmed the correctness and validity of the application of this mathematical apparatus for the set tasks