随着飞行流量激增，空情态势日益复杂，管制员工作负荷与压力越来越大，陆空通话指挥极易出现“错、忘、漏”等情况，造成严重的“人为”安全隐患。如何在现有管制规则和技术条件下减小陆空通话出错的概率，降低工作负荷，提升安全水平成为民航空管亟待解决的突出问题之一。本课题针对陆空通话过程中出现的问题，开展空管语音语义序列标注方法、基于深度学习的空管语音识别、基于深度学习的空管指令理解、空管指令安全实时监控标准与方法等问题的深入研究，以解决陆空通话规范性、飞行员复诵一致性、飞行行为与管制指令一致性和管制指令安全性4类管制通话潜在风险。预期可以建立研究用空管语音识别和语义理解的基线样本数据库；补全空管指挥“人在回路”中最后“未数字化”的一环，给空管双岗值班机制增加一个智能化的安全监视能力，以减少管制人为因素差错，降低工作负荷与压力；为构建具有自主知识产权智能空管系统提供必要的理论技术支撑与实验依据。

With the rapid increase of flights and complexity of air traffic situation, air traffic controllers (ATCOs) are suffering heavier workload and pressure. The wrong, forgotten, missed and other circumstances are inclined to appear during radiotelephony communication in air traffic control, which may cause serious man-made safety risks. Therefore, reducing communication-related problems, relieving workload of ATCOs and improving the overall aviation safety level under the current regulations and technical conditions have become an urgent issue of air traffic control in CAAC. In this project, we focus on four key technologies research based on deep learning , including ATC utterance and semantic annotation method, ATC speech recognition, ATC instruction interpretation and ATC instruction safety monitoring . Through these studies, four categories potential risks of controller-pilot communication can be resolved, including radiotelephone normative, pilot recitation consistency, flight behavior and regulatory compliance ,and control instruction safety. Further, this study can establish a baseline sample database for research on ATC speech recognition and semantic understanding. It can also make up the last un-digitized human-in-loop part of the ATC command and add an intelligent safety monitoring assistant to the dual duty mechanism . It provides the necessary theoretical and technical support and experimental basis for building an intelligent ATC system with independent intellectual property rights.