Research On CGF-Oriented Adversarial Intent Recognition Modeling Method And Application Based On Non-cooperative Game And Learning Theory

Computer Generated Force(CGF)is one of the frontier and key technologies in combat simulation.In order to solve the problem of the insufficient in CGF confrontation ability and unrealistic behavior in the existing simulation system,it is necessary to improve the cognitive level of CGF.Intention recognition is an important cognitive behavior.CGF with intent recognition ability can use the acquired intelligence information to judge the opponent's intention and formulate effective countermeasures.At the same time,in the confrontation field and complex battlefield environment,more attention is paid to how to expose opponent's intentions and plans with higher efficiency by the subjective initiative,so as to obtain credible battlefield intelligence.Therefore,CGF-oriented adversarial intent recognition based on non-cooperative game and learning theory is of great significance for improving the intelligence,humanity and confrontation of CGF.This paper first introduces the research background,siginificance,state of the art of theories and applications.Then,general research framework of intent recognition and the non-cooperative game theory and confrontational intent recognition design are introduced.On this basis,this paper studies the deceptive behaviors in the confrontational environment,and the learning behavior modeling and non-cooperative game method to identify the opponent's deceptive behavior.(1)Deception-based path planning methodBased on the deceptive behavior of the opponent in the environment,this paper analyzes the formal representation of the path planning based on deception,and starts from the different target probability values of each location node based on the model-based intent identification,and quantifies the simulation and dissimulation of each step.In the end,the deceptive path planning method based on explicit modeling and data-driven is proposed.In the experiment,by comparing with the non-spoofing path,the deception effect of different deception strategies and different deception path planning paths methods is demonstrated.(2)Deceptive behavior model method based on maximum entropy inverse reinforcement learningIn the confrontational environment,the opponent uses the deceptive behavior to hide the real target.In order to enhance the recognition effect,the identification party uses the maximum entropy inverse reinforcement learning to establish the deceived behavior model of the opponent.The candidate feature values are proposed from the identified task,and the statistical methods are used to filter and combine feature values to reduce the IRL calculation and improve the IRL effect.Experiments show that using the IRL learning model in the intent recognition can greatly improve the real intention recognition probability of the opponent.(3)Modeling method of confrontation intent recognition based on non-cooperative gameThe non-cooperative game-based confrontation intent recognition modeling method is used to solve the optimal path of the identification party,forcing the opponent to expose the real intention early,and the identification party further enhances the probability of identifying the real target.This paper studies the non-cooperative game utility function based on IRL,solves the Nash equilibrium solution by using the linear programming method,and solves the optimal path blocking with the greedy algorithm.Experiments show that the non-cooperative game based confrontation intent recognition modeling method can further improve the recognition of real target probability.Finally,we summarize the research work of the thesis and give the research questions and theoretical methods that need to be focused on in the future.