Research On Digital Waterway Data Enhancement Technology Based On Deep Learning

In recent years,with the continuous advancement of the construction of inland waterways,the channel status data collected through various sensing methods has become more and more abundant,which has improved the efficiency and service level of channel management.However,due to the limitations of sensing facilities,transmission networks and software systems,the digital waterway sensing data such as ship AIS,water level data,navigation beacon data,etc.are missing,errors or lack of precision,which not only affects the accurate grasp of the real-time state of the channel,but also It will cause great troubles and obstacles to followup channel big data analysis and intelligent applications.In response to this problem,this thesis takes the inland waterway perception data as the object,uses deep learning methods to study their data enhancement methods and models,to fill in the missing data,and lay the foundation for intelligent prediction based on these data.The main research works completed in this thesis includes:(1)Preprocessing of digital channel sensing data.According to the requirements of building a data enhancement model,according to the characteristics of ship AIS,navigation aids data and water level data,their data feature dimensions and time dimensions are screened to form a data set for deep learning training and testing.In view of the spatial correlation of the ship’s AIS trajectory,DBSCAN is used for cluster analysis,and the AIS data set is further processed in sections to capture the behavior characteristics of the ship in different segments more accurately and improve the accuracy of the data enhancement model.(2)Attribute filling model of digital waterway perception data.Aiming at the missing phenomenon of AIS attribute dimension,this paper studies the data attribute filling model based on GAIN,and realizes the filling of AIS attribute missing value by using the game of generating confrontation network.The experimental results show that the GAIN-based model has higher accuracy than the traditional machine learning model.(3)The time dimension enhancement model of digital waterway perception data.Aiming at the lack of ship AIS data,navigation aids data and water level data in the time dimension,their time dimension enhancement models are respectively proposed.The attention mechanismbased CNN-GRU model is used for AIS data enhancement,the CNN-GRU-based model is used for water level data enhancement,and the attention mechanism-based GRU model is used for navigation mark data enhancement.Experimental results show that the proposed model has higher accuracy than traditional machine learning models.Based on the characteristics of digital waterway data,this thesis studies the data enhancement model based on deep learning.By filling the missing values of attribute dimension and time dimension,a more complete data set can be obtained,which is very helpful for channel big data analysis and intelligent prediction based on big data.It is a useful exploration of the application of artificial intelligence in the field of digital channel,It not only has theoretical significance,but also has good application value.