为了预测航空器滑行预计到达时间(Estimated time of arrival,ETA),减少场面冲突,提高机场运行效率,本文使用卡尔曼滤波算法对场面历史轨迹数据进行预处理。为了衡量轨迹样本间的距离,综合三类特征用于机场场面历史轨迹数据聚类。特征...为了预测航空器滑行预计到达时间(Estimated time of arrival,ETA),减少场面冲突,提高机场运行效率,本文使用卡尔曼滤波算法对场面历史轨迹数据进行预处理。为了衡量轨迹样本间的距离,综合三类特征用于机场场面历史轨迹数据聚类。特征包含航空器滑行时段和场面航空器数量,以及参考动态时间规整(Dynamic time warping,DTW)算法提取的轨迹差异度特征。将两个样本特征的欧式距离作为样本间的相似度量;基于均差最大原则确定初始聚类中心,使用K-means算法对样本进行聚类,根据待规划航空器的所处时段和场面航空器数量选择匹配度最高的类簇,将其聚类中心样本的轨迹序列和塔台规划的静态路径相结合预测航空器滑行ETA。通过将实际轨迹数据与预测的滑行ETA进行对比分析,证明了本文预测航空器滑行ETA的准确性。展开更多
Multiple Uninhabited Aerial Vehicles (multi-UAVs) coordinated trajectory replanning is one of the most complicated global optimum problems in multi-UAVs coordinated control. Based on the construction of the basic mode...Multiple Uninhabited Aerial Vehicles (multi-UAVs) coordinated trajectory replanning is one of the most complicated global optimum problems in multi-UAVs coordinated control. Based on the construction of the basic model of multi-UAVs coordinated trajectory replanning, which includes problem description, threat modeling, constraint conditions, coordinated function and coordination mechanism, a novel Max-Min adaptive Ant Colony Optimization (ACO) approach is presented in detail. In view of the characteristics of multi-UAVs coordinated trajectory replanning in dynamic and uncertain environments, the minimum and maximum pheromone trails in ACO are set to enhance the searching capability, and the point pheromone is adopted to achieve the collision avoidance between UAVs at the trajectory planner layer. Considering the simultaneous arrival and the air-space collision avoidance, an Estimated Time of Arrival (ETA) is decided first. Then the trajectory and flight velocity of each UAV are determined. Simulation experiments are performed under the complicated combating environment containing some static threats and popup threats. The results demonstrate the feasibility and the effectiveness of the proposed approach.展开更多
针对即时配送“最后一公里”的问题,综合利用订单取送点、即时配送骑手历史时空轨迹、兴趣面(area of interest,AOI)空间范围与门禁位置等数据,精确预估AOI内部各兴趣点(point of interest,POI)到相应可通行门禁点的时间、距离及路径。...针对即时配送“最后一公里”的问题,综合利用订单取送点、即时配送骑手历史时空轨迹、兴趣面(area of interest,AOI)空间范围与门禁位置等数据,精确预估AOI内部各兴趣点(point of interest,POI)到相应可通行门禁点的时间、距离及路径。在此基础上设计了配套的调用选优策略,获得最优的末端指引方案,以有效提高即时配送路径质量及时间距离预估准确性。展开更多
门控循环单元(gated recurrent unit, GRU)是一种有代表性的深度神经网络,它在众多序列学习任务中达到了国际领先的水平.然而,在门控循环单元的每个时间步之间,输入信息与隐含状态信息缺乏交互,这对更好地挖掘上下文语义信息带来了挑战...门控循环单元(gated recurrent unit, GRU)是一种有代表性的深度神经网络,它在众多序列学习任务中达到了国际领先的水平.然而,在门控循环单元的每个时间步之间,输入信息与隐含状态信息缺乏交互,这对更好地挖掘上下文语义信息带来了挑战.针对这个问题,本文提出了一个新颖的序列学习通用的语义特征提取模型:交互门控循环单元(interactive gated recurrent unit, InterGRU),可以让输入与隐含状态向量在各时间步间进行多轮充分的交互.并且,在到达时间估计(estimated time of arrival, ETA)这个有代表性、有挑战的时空序列预测任务上,本文提出了一套基于交互门控循环单元的深度学习框架(InterGRU-ETA).本文在来自滴滴出行平台真实场景下的海量数据集上充分地实验验证了InterGRU-ETA.结果表明,我们的框架在预测准确率上优于目前国际上最先进的方法.这反映了交互门控循环单元在捕获序列语义信息上的性能优势和广阔前景.展开更多
文摘为了预测航空器滑行预计到达时间(Estimated time of arrival,ETA),减少场面冲突,提高机场运行效率,本文使用卡尔曼滤波算法对场面历史轨迹数据进行预处理。为了衡量轨迹样本间的距离,综合三类特征用于机场场面历史轨迹数据聚类。特征包含航空器滑行时段和场面航空器数量,以及参考动态时间规整(Dynamic time warping,DTW)算法提取的轨迹差异度特征。将两个样本特征的欧式距离作为样本间的相似度量;基于均差最大原则确定初始聚类中心,使用K-means算法对样本进行聚类,根据待规划航空器的所处时段和场面航空器数量选择匹配度最高的类簇,将其聚类中心样本的轨迹序列和塔台规划的静态路径相结合预测航空器滑行ETA。通过将实际轨迹数据与预测的滑行ETA进行对比分析,证明了本文预测航空器滑行ETA的准确性。
基金supported by the Natural Science Foundation of China (Grant no.60604009)Aeronautical Science Foundation of China (Grant no.2006ZC51039,Beijing NOVA Program Foundation of China (Grant no.2007A017)+1 种基金Open Fund of the Provincial Key Laboratory for Information Processing Technology,Suzhou University (Grant no KJS0821)"New Scientific Star in Blue Sky"Talent Program of Beihang University of China
文摘Multiple Uninhabited Aerial Vehicles (multi-UAVs) coordinated trajectory replanning is one of the most complicated global optimum problems in multi-UAVs coordinated control. Based on the construction of the basic model of multi-UAVs coordinated trajectory replanning, which includes problem description, threat modeling, constraint conditions, coordinated function and coordination mechanism, a novel Max-Min adaptive Ant Colony Optimization (ACO) approach is presented in detail. In view of the characteristics of multi-UAVs coordinated trajectory replanning in dynamic and uncertain environments, the minimum and maximum pheromone trails in ACO are set to enhance the searching capability, and the point pheromone is adopted to achieve the collision avoidance between UAVs at the trajectory planner layer. Considering the simultaneous arrival and the air-space collision avoidance, an Estimated Time of Arrival (ETA) is decided first. Then the trajectory and flight velocity of each UAV are determined. Simulation experiments are performed under the complicated combating environment containing some static threats and popup threats. The results demonstrate the feasibility and the effectiveness of the proposed approach.
文摘针对即时配送“最后一公里”的问题,综合利用订单取送点、即时配送骑手历史时空轨迹、兴趣面(area of interest,AOI)空间范围与门禁位置等数据,精确预估AOI内部各兴趣点(point of interest,POI)到相应可通行门禁点的时间、距离及路径。在此基础上设计了配套的调用选优策略,获得最优的末端指引方案,以有效提高即时配送路径质量及时间距离预估准确性。
文摘门控循环单元(gated recurrent unit, GRU)是一种有代表性的深度神经网络,它在众多序列学习任务中达到了国际领先的水平.然而,在门控循环单元的每个时间步之间,输入信息与隐含状态信息缺乏交互,这对更好地挖掘上下文语义信息带来了挑战.针对这个问题,本文提出了一个新颖的序列学习通用的语义特征提取模型:交互门控循环单元(interactive gated recurrent unit, InterGRU),可以让输入与隐含状态向量在各时间步间进行多轮充分的交互.并且,在到达时间估计(estimated time of arrival, ETA)这个有代表性、有挑战的时空序列预测任务上,本文提出了一套基于交互门控循环单元的深度学习框架(InterGRU-ETA).本文在来自滴滴出行平台真实场景下的海量数据集上充分地实验验证了InterGRU-ETA.结果表明,我们的框架在预测准确率上优于目前国际上最先进的方法.这反映了交互门控循环单元在捕获序列语义信息上的性能优势和广阔前景.