为了提高效率,降低培训成本并推广使用计算机来取代管制模拟机中的飞行员席位,采用集成学习的策略来生成飞行员复诵指令。选用5个大规模预训练语言模型进行微调,并使用K折交叉验证来筛选出性能较好的4个模型作为基础模型来构建集成学习...为了提高效率,降低培训成本并推广使用计算机来取代管制模拟机中的飞行员席位,采用集成学习的策略来生成飞行员复诵指令。选用5个大规模预训练语言模型进行微调,并使用K折交叉验证来筛选出性能较好的4个模型作为基础模型来构建集成学习模型。所构建的集成学习模型在管制指令数据集上取得在本领域中的最优效果。在通用的ROUGE(recall-oriented understudy for gisting evaluation)评价标准中,取得R_(OUGE-1)=0.998,R_(OUGE-2)=0.995,R_(OUGE-L)=0.998的最新效果。其中,R_(OUGE-1)关注参考文本与生成文本之间单个单词的匹配度,R_(OUGE-2)则关注两个连续单词的匹配度,R_(OUGE-L)则关注最长公共子序列的匹配度。为了克服通用指标在本领域的局限性,更准确地评估模型性能,针对生成的复诵指令提出一套基于关键词的评价标准。该评价指标准基于管制文本分词后的结果计算各个关键词指标来评估模型的效果。在基于关键词的评价标准下,所构建模型取得整体准确率为0.987的最优效果,对航空器呼号的复诵准确率达到0.998。展开更多
With the increasing civil aviation passengers and the rapid development of aviation logistics,the study on remotely piloted operation(RPO)mode has received extensive attention.RPO mode constructs the piloting decision...With the increasing civil aviation passengers and the rapid development of aviation logistics,the study on remotely piloted operation(RPO)mode has received extensive attention.RPO mode constructs the piloting decisionmaking mode which involves the tripartite collaboration among airborne automatic/autonomous system,remote ground-based crews and air traffic control.In this paper,we describe the organizing architecture for commercial remotely piloted aircraft(CRPA)system and its components.Compared with the current operation mode,the new air-ground collaborative decision-making mode has been established with six different situations based on the type of the flight and the condition of the remote pilot.Taking airport surface operation as an experimental example,we model the airport surface operation process and compare the advantages and disadvantages between RPO mode and the current dual-pilot mode from the perspectives of time and operation coverage,and draw conclusions that RPO mode can basically cover the flight operations of the dual-pilot,improve the accuracy of pilot operations and greatly reduce response time by 48%in pre-flight inspection.The above research would be the foundation for the RPO development of commercial aircraft in China.展开更多
文摘为了提高效率,降低培训成本并推广使用计算机来取代管制模拟机中的飞行员席位,采用集成学习的策略来生成飞行员复诵指令。选用5个大规模预训练语言模型进行微调,并使用K折交叉验证来筛选出性能较好的4个模型作为基础模型来构建集成学习模型。所构建的集成学习模型在管制指令数据集上取得在本领域中的最优效果。在通用的ROUGE(recall-oriented understudy for gisting evaluation)评价标准中,取得R_(OUGE-1)=0.998,R_(OUGE-2)=0.995,R_(OUGE-L)=0.998的最新效果。其中,R_(OUGE-1)关注参考文本与生成文本之间单个单词的匹配度,R_(OUGE-2)则关注两个连续单词的匹配度,R_(OUGE-L)则关注最长公共子序列的匹配度。为了克服通用指标在本领域的局限性,更准确地评估模型性能,针对生成的复诵指令提出一套基于关键词的评价标准。该评价指标准基于管制文本分词后的结果计算各个关键词指标来评估模型的效果。在基于关键词的评价标准下,所构建模型取得整体准确率为0.987的最优效果,对航空器呼号的复诵准确率达到0.998。
基金supported by the National Program on Key Basic Research Project (No. 2014CB744903)the National Natural Science Foundation of China(Nos. 61973212,61673270)+3 种基金the Shanghai Industrial Strengthening Project (No. GYQJ-2017-5-08)the Shanghai Science and Technology Committee Research Project (No. 17DZ1204304)the Civil Aviation Pre-Research ProjectsShanghai Engineering Research Center of Civil Aircraft Flight Testing.
文摘With the increasing civil aviation passengers and the rapid development of aviation logistics,the study on remotely piloted operation(RPO)mode has received extensive attention.RPO mode constructs the piloting decisionmaking mode which involves the tripartite collaboration among airborne automatic/autonomous system,remote ground-based crews and air traffic control.In this paper,we describe the organizing architecture for commercial remotely piloted aircraft(CRPA)system and its components.Compared with the current operation mode,the new air-ground collaborative decision-making mode has been established with six different situations based on the type of the flight and the condition of the remote pilot.Taking airport surface operation as an experimental example,we model the airport surface operation process and compare the advantages and disadvantages between RPO mode and the current dual-pilot mode from the perspectives of time and operation coverage,and draw conclusions that RPO mode can basically cover the flight operations of the dual-pilot,improve the accuracy of pilot operations and greatly reduce response time by 48%in pre-flight inspection.The above research would be the foundation for the RPO development of commercial aircraft in China.
