Research On Disintegration Of Heterogeneous Combat Network Under Incomplete Information

The rapid development of military technology has prompted different types of com-bat entities on the battlefield to break the limits of the operational domains,and a huge combat network emerge through complex interactions between different combat entities.This makes the military confrontation is no longer limited to confrontation centered on equipment,but reflects the confrontation between the systems of the two belligerents.The winning mechanism of war is thus transformed into the disintegration of the enemy’s com-bat system of systems(CSo S).CSo S disintegration focus on the key equipment to make structural damage to the enemy’s CSo S,which can achieve system victory.However,un-der the conditions of information,the battlefield is increasingly complex,reflecting great uncertainty.The complex battlefield leads to the lack of completeness and authenticity of enemy information that can be obtained,which greatly affects command and decision-making and makes it very difficult to dismantle the CSo S.To solve this problem,this paper abstracts the enemy CSo S as a heterogeneous com-bat network(HCN),and proposes a framework for the HCN disintegration problem under incomplete information conditions.The research of this paper is carried out according to the idea of ”heterogeneous combat network modeling → network reconstruction → net-work disintegration”.The key technologies involved in this paper include: heterogeneous network-based CSo S modeling,HCN link prediction based on network representation learning,and HCN disintegration based on reinforcement learning.We perform link pre-diction for HCNs by fully mining the semantic and topology information to achieve the reconstruction of HCN.Further,we study the disintegration strategy of the HCN based on reinforcement learning.The main research work and innovations of this paper include:(1)We proposed a networked modeling method for CSo S based on heterogeneous networkIn modern warfare,different types of combat entities form a CSo S through complex interactions,presenting large scale and complex structure characteristics.To this end,this paper proposes a network modeling method for CSo S based on heterogeneous net-works.First,according to the functions exerted by combat entities in the battlefield to model different types of entities as nodes in the network? Then,combing the interactions between different types of nodes,and define the interaction between different types of nodes normatively,abstracting them as heterogeneous links.Finally,a HCN model is constructed.The model reflects the multiple and complex interactions between different types of entities,and portrays the CSo S.(2)We proposed a framework for HCN disintegration under incomplete informationWe combine the characteristics of HCNs to propose a capability assessment index based on kill web,which is used as the objective function of network disintegration.Fur-ther,the mathematical optimization model of HCN disintegration is defined.On this basis,a framework for HCN disintegration under incomplete information is proposed.Firstly,the enemy CSo S is modeled as a heterogeneous network.Then,the HCN is re-constructed by link prediction.Finally,the disintegration strategy of the HCN is studied.The framework can effectively support command decisions on the battlefield and improve the effectiveness of disintegration on enemy combat networks.(3)We proposed HCNs link prediction based on representation learningWe use meta-path-based random walk to mine the structural and semantic informa-tion embedded in the HCN.And the random walk data is inputted into the network rep-resentation learning model as training data to obtain the node embedding.Further,a link prediction framework based on representation learning is proposed.At first,the node embedding is obtained through network representation learning.Then,calculate the sim-ilarity between node pairs based on the node embedding.Finally,perform link prediction for the HCN based on the node similarity.(4)We proposed HCN disintegration strategy based on reinforcement learningTo solve the HCN disintegration problem with cost constraints,a framework for solving this problem is proposed based on reinforcement learning.We firstly proposed a framework for solving network disintegration strategy based on the Q-Table,which is only applicable to solving small-scale disintegration problems.To this end,we further propose a framework for solving network disintegration strategies based on Double Deep Q-Learning,which enables to solve larger-scale network disintegration problems by in-troducing neural networks.It is experimentally demonstrated that the framework can effectively solve the HCN disintegration problem with cost constraints.