基于预训练语言模型的管制信息抽取方法

Air traffic control information extraction method based on pre-trained language models

在空中交通管理中,管制员使用管制指令调节航空器状态,飞行员通过复诵指令进行确认。管制指令的正确理解对飞行安全具有重大意义。提出一种新的管制信息抽取方法,即基于语言模型的预训练和微调,通过迁移学习实现小样本管制信息抽取。该方法在训练数据量降低时,仍能实现准确率的提升。仿真结果表明,新模型对管制信息抽取的准确率不低于98%,可以有效提取管制指令中的关键信息。该方法可提升空管系统的智慧化程度,辅助管制员理解管制指令内容,支撑飞行冲突检测,保障航空运输安全。

In air traffic management,the controller uses the control instruction to adjust the aircraft status,and the pilot confirms by repeating the control instruction.The correct understanding of control instruction is of great significance to flight safety.This paper proposes a new method of air traffic control information extraction,which is based on pre-training and fine tuning of the pre-training language model.It uses transfer learning to extract regulatory information under the condition of small samples.This method can not only reduce the cost of training data annotation,but also improve the accuracy of information extraction.The simulation results show that the accuracy of the new model is not less than 98%,and the key information in the control instructions can be extracted effectively.This method can improve the intelligence of air traffic control system,assist controllers to understand the contents of control instructions,support flight conflict detection,and ensure air transport safety.