A novel method of heterogeneous combat network disintegration based on deep reinforcement learning

Abstract

Modern war is highly dependent on intelligent, unmanned combat systems. Since many intelligent, unmanned combat systems have network attributes, it is meaningful to research combat systems from the perspective of complex network. Heterogeneous network provides a suitable model to describe real combat network. Previous studies of combat network only concentrate on homogeneous networks. However, on the real battlefield, military networks are composed of a large number of heterogeneous nodes and edges with different functions. In the paper, a superior, intelligent, heterogeneous combat network disintegration strategy (HDGED) are obtained by DQN, which embeds heterogeneous networks into a low-dimensional representation vector as input, rather than ignore the differences of the nodes and their connections. A method of heterogeneous graph embedding is first introduced, which adopts type encoding and aggregation. Besides, a normalized combat capability index was designed, which could assess the performance of the dynamic heterogeneous combat networks. On this basis, HDGED was experimented on networks with uneven node combat capabilities and the results show that HDGED has improved disintegration effectiveness for heterogeneous networks of different sizes compared with traditional methods. Our work provides a new approach to realize the disintegration of heterogeneous combat networks by deep reinforcement learning, which is of great significance for optimizing the command operation process, and deserves further study.