Research On Key Technology Of Human-Machine Dialogue System For Radiotelephony Communications

Radiotelephony communications,which has the characteristics of naturalness,flexibility and convenience,is the basic working mode in air traffic control to ensure the safe and efficient operation of air traffic.The harmonious communication between the controller and the pilot can avoid the harm caused by human factors.It is of great practical significance to improve the efficiency and fidelity of air traffic controller training by using computer instead of pilot seat to response.Firstly,in this paper,the corresponding data sets were created according to different natural language processing task after conversation structure and linguistic features of radiotelephony communications were analyzed,including: radiotelephony communications dictionary,radiotelephony communications instruction set,radiotelephony communications corpus data set.The specific word embeddings of radiotelephony communications was generated by radiotelephony communications corpus data set which can effectively characterize ATC instruction semantics.Secondly,in view of the uniqueness and specialty of radiotelephony communications and the lack of existing marking materials,special evaluation methods for the word embeddings of radiotelephony communications were proposed.The conceptual classification method based on k-means was used to evaluate word embeddings internally,which proved the functions of word embeddings in representing semantics and distinguishing word categories simply.The similarity calculation model based on siamese network was used to evaluate the validity of word embeddings for natural language processing downstream tasks with similar principles.Finally,a dialogue model used for radiotelephony communications based on seq2 seq network framework was established and trained by the word embeddings of radiotelephony communications and evaluated from four indicators,including accuracy and training duration and others.Aiming at the shortcomings of the basic dialogue model,such as response of numerical instruction was prone to error and instruction data coverage was incomplete,an improved dialogue model combining deep learning and grammar rules of radiotelephony communication was proposed,which further improved the effect of the model.The response accuracy increased from93.1% to 95.7%,and the training duration shortened by 1 hour and 33 minutes.The improved model satisfied the need to input the controller’s instructions to get the correct pilot response.The dialogue system between controller and pilot can realize the automation of pilot position in the controller training,and is worthy of further research as a direction to realize the intellectualization of ATC.In this paper,an improved dialogue system was built for the special expressions of radiotelephony communications which showed good effect in the stage of approach and departure,and had a high accuracy rate of response.