The paper proposed approach to estimation the resilience of cyber-physical systems, as well as a method for their reconfiguration to neutralize the negative effects of structural attacks. The proposed method is applied to systems modeled by graphs, each vertex of which is associated with attributes - types of devices. The functioning of such systems is determined by the path on the graph, passing through the vertices of a given type. The reconfiguration method based on the graph artificial neural network (ANN) aims at increasing the number of working paths without the need to add new edges. The ANN model was trained on a synthetic dataset composed of random graphs whose vertex types were specified according to the betweenness centrality metric