A Study Of Operation Loop Recommendation Methods For Kill Web In The Mosaic War

With the new round explosive development of computing power,network communication,artificial intelligence and other technologies,computer-aided human decision-making has also changed from “aided computing oriented” to “Aided cognitive enhancement oriented”.In the military field,decision-centric warfare has become one of the cores of the new generation operational concept: Mosaic War.In a new military landscape in which the equipment generation gap among major countries is gradually narrowing,how to make more efficient and accurate decisions has become a new point for military competition among countries.Around the characteristics of Mosaic War,focusing on the basic combat form of “net-killing” in the future,based on the operation loop theory,this dissertation carries out a series of methods research of operation loop recommendation(OLR)among kill web in the Mosaic war,to provide theoretical,method and algorithm reference for the development and construction of our military’s intelligent C2 and decision.First,the research framework of OLR for kill web in the Mosaic War is constructed.Based on the theory of “5W1H”,this dissertation makes a detailed analysis of the OLR problem for kill web in the Mosaic War from the six dimensions of “what”,“why”,“essence”,“application scenario”,“demand”,and “factor”,clarifies the boundary,target,demand,constraint and other factors of the problem,and confirms the research route.Second,considering the heterogeneity of nodes and edges in the kill web of Mosaic War,the modeling method of OLR problem based on heterogeneous network is studied.Based on heterogeneous network theory,the modeling method of heterogeneous nodes,network mode and meta-path of kill web is proposed.Combined with the military maturity theory knowledge,the calculation model of edge quality in kill web is defined.Then,according to the demands and considerations of the problem,the OLR programming model is designed,which takes the quality of the operation loop as the effectiveness objective,the time utility as the constraint,and the target value as the weighting factor.Then,based on the open information,the unified case study scenarios and data are designed,the kill web model of the case is constructed by using the research method in this part,and the relevant merits of the example operation loop are calculated.Third,for the linear evaluation case,the OLR method based on accurate algorithm is studied.The heterogeneous kill web was transformed into a normal weighted network.Considering the effective remaining strike time of the target,the OLR problem was transformed into a shortest path problem with time constraint,and the OLR programming model based on graph was constructed.Based on the idea of branch and bound,an accurate algorithm of OLR with time constraint is designed.The case study shows that the algorithm can obtain the precise optimal solution for the OLR problem of small scale,and the consumed time of OLR optimization without time constraint is much less than that with time constraint.Fourth,for the nonlinear evaluation case,the OLR method based on integrated improved ant colony algorithm(ACO)is studied.Aiming at the drawbacks of slow convergence in the early stage,being greatly affected by parameters,and being easy to fall into local optimal of the original ACO,three improvement strategies were proposed respectively: pheromone initialization based on edge weight information,adaptive optimization of ACO parameters based on differential evolution,and improvement of global search ability based on genetic operator.A case study is conducted to analyze and compare the proposed algorithm.It is proved that the proposed algorithm is better than the unintegrated improved ACO without greatly increasing the time consumption,and its effect is significantly improved compared with other classical heuristic algorithms.Also,the proposed algorithm supports the optimization of large-scale problems.Fifth,for the “black box” evaluation case,the OLR method based on deep reinforcement learning(DRL)is studied.A multi-stage planning model of OLR problem was built and mapped to the reinforcement learning(RL)model.The RL elements such as environment,state,action,state transition and reward were defined.Based on Rainbow DRL algorithm,the state vector,neural network structure and algorithm framework and flow of OLR are designed.The case study verifies that the proposed algorithm is better than the traditional DQN(Deep Q Learning)algorithm and the accurate algorithm.The DRL agent can be trained based on the data generated by the “black box”,and can be deployed offline after training,which can almost instantly complete the OLR.Sixth,for the case with multiple decision objectives,the multi-objective optimization method of OLR based on MOACEA/D(multi-objective ant colony evolutionary algorithm/decomposition)is studied.Considering the two objectives of effectiveness and loop closing time of the OLR scheme,a multi-objective programming model of OLR is constructed.Based on the framework of multi-objective evolutionary algorithm/decomposition(MOEA/D),within the idea of ACO,the OLR scheme is constructed by using pheromone to build a new algorithm of MOACEA/D.The case analysis was compared with other classical multi-objective optimization algorithms.It is verified that the proposed improved algorithm is superior to other algorithms in the hypervolume index.Seventh,for the case of losing global C2 and decision capability,the OLR method based on autonomous cooperation-confrontation rules is studied.The agent model of the equipment system and the interaction model between the systems are constructed.Two kinds of behavior rules of the equipment agent are designed: the “cooperationconfrontation” rule based on the operation loop,and the movement rule based on the cluster movement.In the case analysis,the autonomous behavior rules of the equipment agent are simulated and verified based on Net Logo.The effectiveness of the proposed behavior rules is verified by showing the systems dynamic confrontation process of two camps.In addition,the influence of different deployment positions of the equipment on the confrontation results and the catastrophe point of the communication capability of certain equipment are also analyzed.It shows that the constructed simulation model can support the researchers to conduct extensive system analysis.