Research On Man-machine Dialogue System Of Control Simulator Based On Seq2Seq Model

With the improvement of the national economic level,China’s civil aviation began to develop rapidly,the number of flights increased year by year,and the air traffic flow increased steadily.Therefore,the training requirements for air traffic controllers are also increasing year by year.At present,the training of air traffic controllers in China is mainly based on air traffic control simulator,and the use of computer instead of pilot for command response is an effective way to improve the training of controllers and the utilization of simulator.This paper makes an in-depth study on the rationality of the development of man-machine dialogue system of control simulator,comprehensively analyzes the problems of control command recognition and response command synthesis in land air call,puts forward an improved seq2seq land air call dialogue model,and builds a visual platform of dialogue system,which lays a foundation for the intellectualization of simulator.Firstly,starting from the Air traffic radio communication Terminology Standard,this paper analyzes the unique dialogue structure and language characteristics of land air communication,and summarizes the standardized dialogue structure of various control terms such as approach control and regional control.Then,according to the structural characteristics of land air conversation,a land air conversation corpus of proper nouns such as flight,letter,altitude,waypoint and instruction is constructed,and the preprocessing work such as Chinese word segmentation and part of speech tagging is completed on this basis.Secondly,word2 vec training tool is selected to train the data set.Compared with sparse word vector representation method,it reduces the dimension of word vector and the amount of network calculation.At the same time,word2 vec can also map words with similar semantics into the corresponding vector space,so that the word vector has semantic characteristics,which can effectively express the instruction meaning in land air communication.Then,the seq2seq network architecture based on attention mechanism is selected to build the land air call dialogue model.On this basis,the land air call word vector is used for training.The experimental results are statistically compared by adjusting the learning rate and network layer parameters,and the training results are analyzed with the training time and accuracy as the measurement indicators.Aiming at the defects in the training process,an improved land air conversation model is proposed.The training results show that the accuracy of the model is improved to 93% and the time is shortened to 4h10 min,which meets the needs of computer instead of pilot to realize real-time response to control instructions.Finally,the dialogue system of land air call command response is designed to display the research results of this paper.The system consists of login interface and command response dialogue interface.The system provides research support for the completion of intelligent controller computer dialogue simulation training system instead of pilots in the future.