近距空战训练中的智能虚拟对手决策与导引方法

针对近距空战训练中智能虚拟对手攻防博弈的自主决策与占位导引问题，提出了基于动态贝叶斯网络(DBN)和约束梯度法的智能虚拟对手决策和导引一体化方法。结合空间占位态势、火控攻击区和机动动作识别结果等信息，建立近距空战决策动态贝叶斯网络模型，实现根据战场动态环境变化的占位导引指标决策。针对在线识别的各类目标机动动作，建立轨迹预测模型，实现目标轨迹的实时预测。根据占位导引指标和目标预测轨迹，考虑飞行性能约束，采用约束梯度法计算智能虚拟对手的优化占位导引量。实现了近距空战智能虚拟对手空间占位决策与导引量计算的无缝结合。近距空战仿真实验结果表明:所提方法能够实现智能虚拟对手的合理化自主决策和占位导引，克服了传统方法实现机动动作方式固化的问题，具备较好的实时性和优化性。 

Intelligent virtual opponent decision making and guidance method in short-range air combat training

To train pilots' short-range air combat skills, the traditional way based on flight simulation technology is to have multiple pilots operate multiple fighter simulators at the same time. If an intelligent virtual opponent is used to assist pilots in confrontation training, not only could the normal training process without other pilots be guaranteed, but the training cost could also be reduced to a great extent. In this paper, an integrated method based on dynamic Bayesian network (DBN) and constrained gradient algorithm is proposed to realize autonomous decision making and space occupancy guidance for intelligent virtual opponents in the attack and defense game during short-range air combat training. A dynamic Bayesian network model for short-range air combat decision making is established in combination with the space occupying situation, the fire control attack area and the identification results of maneuvering actions. This model realizes an intelligent selection of occupancy guidance index in accordance with the dynamic battlefield environment. A target trajectory prediction model is built for each type of maneuvers identified online to obtain the real-time prediction of the target trajectory. With the occupancy guidance index, target trajectory predication, and the flight performance constraints in consideration, a constraint gradient method is used to calculate the optimal occupancy guidance quantity of the intelligent virtual opponent. Thus, a seamless combination of space occupancy decision and guidance quantity calculation for intelligent virtual opponent is achieved. The simulation results of short-range air combat show that the proposed method can realize rational autonomous decision making and space occupancy guidance for intelligent virtual opponent, overcome the problem of solidifying the maneuver mode in traditional methods, and thus have better real time and optimization performance.