Space-based AIS Data Mining Based On Machine Learning Method

Building a maritime power is an important part of building a socialist modernization country in an all-round way.As a basic strategic resource,ocean big data is an important symbol to measure a country’s marine regulatory capacity.With the rapid development of information technology,the scale of ocean data represented by satellite ocean remote sensing data shows an explosive growth trend,and ocean research is entering the age of big data.Ocean data collection,management and mining is considered as the key research field by the world’s powerful marine countries.Based on the space-based AIS data collected by Tiantuo-3 satellite developed by National Defense University of Science and Technology and combining with machine learning related algorithms,this paper carries out the research of space-based AIS data mining from three aspects of classification,clustering and regression,we explores the theory and method of data-driven mining for space-based AIS with the purpose to find valuable maritime traffic information from space-based AIS data to enhance maritime situational awareness.The main work of this paper can be concluded as follows:Firstly,on the basis of summarizing the current research status and development trend of AIS data mining,this paper synthetically analyses the unique advantages of space-based AIS data in the field of maritime traffic supervision.Compared with traditional terrestrial AIS and other data sources,space-based AIS has the characteristics of wide coverage,long tracking time and abundant information types,which makes it suitable for large geographical scope and long-term macro-level marine traffic research.Aimed at the practical needs of improving the intelligent level of maritime supervision,this paper points out three directions that can be studied: ship classification,route mining and behavior prediction.A ship classification method based on Random Forest is proposed.According to the characteristics of different vessel types,vessel behavior features are extracted and analyzed on the basis of traditional geometric feature analysis.The effects of different combinations of two main parameters of random forest algorithm on the classification results of five vessel types are discussed.It is found that the extracted features reflect the difference between different vessel types well.The classification accuracy of the algorithm can reach about 80% considering both geometric and behavior features.The importance of features in the classification process is studied,some important features that play a key role in the classification process are found through comparative experiments.An example analysis of one typical vessel with abnormal classification results is conducted,which verifies the effectiveness of the method.。A route mining method based on DBSCAN(Density-Based Spatial Clustering of Applications with Noise)density clustering algorithm is proposed.Two clustering modes are considered,point-based clustering and sub-trajectory-based clustering.For the first clustering mode,the course constraint is added on the basis of classical DBSCAN algorithm and the search zone is improved.According to the clustering experiments on the AIS data of Australian sea area,it is found that the algorithm can effectively extract the hot routes in the target sea area under the appropriate parameters.Through KDE and statistical analysis,we can realize the monitoring of the marine traffic situation of the mining routes.For the second clustering method,a sub-trajectory sample extraction method for space-based AIS data is proposed,which combines DTW distance measurement method to realize the trajectory clustering.Based on the needs of developing Arctic routes,this clustering method is applied to the analysis of AIS data in Arctic region.The clustering results can be used to understand the current situation of Arctic routes and provide information support for the formulation of Arctic strategy.A vessel position prediction method based on LSTM(Long Short-Term Memory)recurrent neural network is proposed.A four-layer LSTM neural network structure is designed and a five-step network training method is used to learn vessel motion patterns from a single ship’s historical data.The prediction results show that the prediction error of the model is controlled to 100 m under normal vessel navigation conditions,however,under the situation of abrupt change of motion state,the prediction error would be large.This method can be used to detect and monitor abnormal ships and to prevent maritime traffic accidents.This paper explores the space-based AIS data mining based on machine learning method,and has made a lot of adaptive improvements at the algorithm level,which has been successfully applied to vessel classification and recognition,route mining and location prediction.This work aims to find out maritime traffic information from space-based AIS data and further enhances the maritime situational awareness and the intelligent level of satellite remote sensing data applications.