Study On Tactics Exploration Method For Equipment Effectiveness Simulation

With the development of technology,the construction of a new equipment gets various technology alternatives.An accurate evaluation on the operation effectiveness of an equipment under different technology alternatives is critical to finalize the design.Engagement-level simulation is a quantitative method to evaluate the effectiveness of an equipment before its construction and acquisition,minimizing the investment risk and also providing feedback to optimize the technology solutions.In the design phase,the combination of technologies leads to a large technological space,consequently,a lot of competent tactics are required to conduct a comprehensive evaluation.Classical tactic modeling methods are ineffective to fully explore the tactical space,and are incapable to support large scale experiments on simulation.The core work and innovation of this thesis include:(1)Tactic exploration framework for equipment effectiveness simulation.This paper analyzes the shortcomings of existing tactical development methods in large-scale tactic exploration and proposes a new tactical exploration framework(TEF).TEF improves the efficiency of tactic development from three aspects: tactic modeling language,tactic exploration algorithm and tactic decision-making model for uncertain environments.First,a domain specific language is proposed for tactic modeling.It provides a high abstract level representation with behavior tree(BT)formalism and implements the automatic generation of tactic scripts.Then,a tactic exploration method based on grammar evolution(GE)algorithm is proposed.It employs the modular behavior tree structures and evolution operators to generate tactic alternatives.Simulation experiments are conducted to test the tactics and feedback the result for GE to optimize tactics.Finally,a tactic decision-making model based on stochastic BT is proposed to operate in an uncertain environment,which is trained with reinforcement learning(RL)to improve combat capability.(2)Tactic modeling language for equipment effectiveness simulation.Tactic representation is a critical part for engagement-level simulation.To improve the abstract level of tactic representation and further support model validation and reuse,a tactic modeling language based on BT is proposed.The advantage of BT formalism as tactic representation is analyzed and an event-driven execution semantics for BT is proposed to optimize the execution of tactic model.Based on the Domain Specific Modeling(DSM),the meta model,syntax and semantics of tactic modeling language is defined.The Object Constraint Language(OCL)is used to define the tactic constraints and validate the tactics.A graphical user interface is designed for experts to edit tactics,which implements automatic generation from model to executable scripts for simulator.On the view of simulation,tactic modeling language provides high abstract representation and code generation,which works as the basis of the TEF.(3)Self-adaptive grammar evolution for tactic exploration.In this paper,a tactical exploration framework based on grammar evolution is proposed.The mapping process from high abstract tactic BTs to binary strings for GE operators is defined.The evolution process is also provided to conduct tactic exploration.Followed,multiple evolution operators are designed to improve the correct rate in generating tactic alternatives,including position wrapped genetic crossover operator,tree based phenotypic operators and preponderant structure combine operator.To improve the performance of GE,self-adaptive methods are proposed to control the exploration process,including a method to coordinate multi-parameters optimization and a Multi-Armed Bandit(MAB)based algorithm to balance the utilization of different operators.The self-adaptive methods dynamically adjust parameters and operators according to their contributions to the exploration.(4)Tactic decision-making model for uncertain environment.A tactic BT is a preset plan and it is unable to deal with undefined situations.Therefore,a tactic decisionmaking model is proposed,which is trained with RL algorithm.First,the BT is extended with stochastic control nodes to change the priorities and constraints between tasks.The stochastic control nodes are modeled with markov decision process(MDP)to make decisions,that change a plan based BT into a tactic decision-making model with both declarative plan and reactive action.Followed,a long short-term memory(LSTM)network is used to deal with partial observed information in battle field and a deep neural network is constructed to make decisions.Finally,a state-of-art RL algorithm called asynchronous advantage actor-critic(A3C)is implemented to train the tactic decision-making model.