Adaptive structure modeling and prediction for swarm unmanned system

In recent years, swarm unmanned systems (SUSs) have become crucial in the military field, both at home and abroad. This has promoted the evolution of the unmanned combat mode from single-platform remote-control to intelligent-swarm combat. SUSs support the cooperative, autonomous, and flexible characteristics of the combat system under uncertain tasks and environments. The overall swarm performance depends on the system and structure among its members and also dynamically evolves with the time and the environment. Thus, new intelligence emerges from the interaction among systems. Starting from the evolution of the SUS structure, this paper proposes the model of a three-layer structure and a relationship involving the data-link layer, the SUS, and the task requirements. The multidimensional spatial relationship model is transformed into a two-dimensional graphical representation model by using a graph neural network; then, the dependency graph of the relationships of the systems and layers of the SUS is constructed. The integral network is classified according to a task-based standard. The recursive neural network algorithm is derived from the intra- and interlayer relationships. The SUS structure is predicted via some examples of training data sets and the attribute labels of task-based nodes. The impact of the system or data layer damage can be evaluated according to the weight parameters of the structure dependence relationship. Finally, the autonomous decision of the SUS from the task to the swarm structure is realized.