Weighted interdependent network disintegration strategy based on Q-learning

Abstract

The problem of network disintegration is one of the core topics in the field of network science. Currently, most of the existing research is based on homogeneous and single-layer networks of nodes. However, the various components of complex systems in the real world are often interdependent, and the cost of attacking different units is diverse, causing the traditional network disintegration method to lack good applicability. This paper establishes a weighted interdependent network (WIN) model, and based on this, a WIN disintegration strategy based on Q-learning is proposed. First, the network nodes are divided into multiple node sets according to the dependencies between the nodes, and the state and action space of Q-learning are determined. Next, the disintegration cost constraints and Q-learning parameters are defined to perform iterative learning. Then, the optimal network disintegration strategy is calculated according to the iterative Q-table. The results show that when the cost sensitivity factor (p) is fixed, DSQ can maintain good results in disintegrating different types of networks under different cost constraints, while the baseline methods have difficulty guaranteeing the disintegration effect in the face of different types of networks. Furthermore, we perform a sensitivity analysis on the p value and find that the effect of most of the baseline methods worsens as the p value increases, while DSQ maintains a good effect.