在进行实时对抗的任务中,对于敌方的动作识别较为困难,需要根据对方的移动轨迹或行为来分析对方的意图,预测其未来目标,构建规划策略库.针对此问题,提出基于数据驱动的多智能体识别算法,该算法首先采用基于自动机的特征提取方法,获得规...在进行实时对抗的任务中,对于敌方的动作识别较为困难,需要根据对方的移动轨迹或行为来分析对方的意图,预测其未来目标,构建规划策略库.针对此问题,提出基于数据驱动的多智能体识别算法,该算法首先采用基于自动机的特征提取方法,获得规划需要的位置和任务信息;然后将规划识别问题转换为多分类问题,并从单智能体角度切入,给出了一种基于极端梯度提升(extreme gradient boosting,XGBoost)的多分类模型;之后,对于多智能体之间可能存在的合作行为,使用无监督学习的一种基于密度对噪声鲁棒的空间聚类算法(density-based spatial clustering of applications with noise,DBSCAN)对多智能体进行分簇,以促进协同合作.对于同簇智能体,构建了一种针对多智能体的多分类模型,完成对多智能体的目标预测.在获悉敌方目标后,提出基于博弈的围捕逼停算法,构建非合作动态博弈模型,通过求解纳什均衡得到应对敌方的最优策略.最后,通过仿真验证了所提出算法的有效性.展开更多
Cooperation is an important mechanism for MultiAgent systems to work effectively. To real-ize it,a distributed plan is introduced in this paper. The constrained relations between subgoals, which have AND/OR relations,...Cooperation is an important mechanism for MultiAgent systems to work effectively. To real-ize it,a distributed plan is introduced in this paper. The constrained relations between subgoals, which have AND/OR relations, can be descrbed in a plan. After negotiating, a steady community will be formed,then the goal will be finished after distributively performing a plan. This kind of cooperation has been realized in the project MasBuilder(MultiAgent Systems Builder).展开更多
文摘在进行实时对抗的任务中,对于敌方的动作识别较为困难,需要根据对方的移动轨迹或行为来分析对方的意图,预测其未来目标,构建规划策略库.针对此问题,提出基于数据驱动的多智能体识别算法,该算法首先采用基于自动机的特征提取方法,获得规划需要的位置和任务信息;然后将规划识别问题转换为多分类问题,并从单智能体角度切入,给出了一种基于极端梯度提升(extreme gradient boosting,XGBoost)的多分类模型;之后,对于多智能体之间可能存在的合作行为,使用无监督学习的一种基于密度对噪声鲁棒的空间聚类算法(density-based spatial clustering of applications with noise,DBSCAN)对多智能体进行分簇,以促进协同合作.对于同簇智能体,构建了一种针对多智能体的多分类模型,完成对多智能体的目标预测.在获悉敌方目标后,提出基于博弈的围捕逼停算法,构建非合作动态博弈模型,通过求解纳什均衡得到应对敌方的最优策略.最后,通过仿真验证了所提出算法的有效性.
文摘Cooperation is an important mechanism for MultiAgent systems to work effectively. To real-ize it,a distributed plan is introduced in this paper. The constrained relations between subgoals, which have AND/OR relations, can be descrbed in a plan. After negotiating, a steady community will be formed,then the goal will be finished after distributively performing a plan. This kind of cooperation has been realized in the project MasBuilder(MultiAgent Systems Builder).
