摘要
针对现有通用机动轨迹需要较长的预规划时间,无法在机载计算平台实时解算的问题,提出一种利用并行计算的方式对通用机动框架进行加速的方法。对现有的MCTS算法叶子节点并行、根节点并行和树并行方式进行分析,结合叶子节点并行和根节点并行方式各自的优点,对每棵搜索树采用叶子节点并行方法,分别利用Pthread和CUDA对并行通用机动框架进行加速,并以筋斗机动为例对加速效果进行测试。实验结果表明:并行通用机动框架不仅性能优于串行框架,而且可大幅缩短机动解算时间。
Aiming at longer pre-planning time for the existing general maneuvering trajectory,it is impossible to solve the problem of real-time solution in the airborne computing platform,and propose a method for accelerating the general maneuvering framework by using parallel computing.Analyzes the leaf node parallel,root node parallel and tree parallel of the existing MCTS algorithm,combined with the advantages of leaf node parallel and root nodes parallel,applied the leaf node parallel method for each search tree,and used the Pthread and Cuda to accelerate the parallel general maneuvering framework,and test the accelerated effect with the example of the maneuvering force.The experimental results show that the parallel maneuvering framework is not only better than the serial frame,but also can shorten the time of the maneuver calculation.
作者
蒋超
王维嘉
王昊
Jiang Chao;Wang Weijia;Wang Hao(Department of Flight Control,AVIC Xi’an Flight Automatic Control Research Institute,Xi’an 710065,China)
出处
《兵工自动化》
2020年第8期25-31,36,共8页
Ordnance Industry Automation
关键词
并行计算
蒙特卡罗树搜索算法
GPU
众核
通用机动框架
parallel computing
Monte Carlo tree search algorithm
GPU
multi-core
general maneuvering frame