Ship Track Prediction Of Inland River Based On Massive AIS Data

With the rapid development of modern inland waterway trade and the increasing traffic volume,the inland river transportation environment has become increasingly complex,and the safety of ships' navigation has become a critical issue.How to predict the future ship trajectory through the historical ship track data,which is the key to achieving safe navigation of inland river ships in maritime management.At the same time,with the development of information technology,the scale of ship AIS data that can be collected continuously increases,the traditional computing platform cannot meet the calculation requirements,and the training speed of the model drops significantly,which seriously affects the real-time performance of track prediction.Therefore,how to improve the ship track prediction efficiency is also a difficult problem that needs to be solved urgently.This paper is based on AIS data generated by navigation of inland river vessels,and proposes a trajectory prediction model based on LSTM and wavelet transform.And based on the Spark distributed computing platform,parallel optimization of trajectory prediction model is realized.The main research work and innovation of this paper are as follows:(1)This paper first analyzes the research status of track prediction and massive data processing,and determines a prediction method based on deep learning and a big data processing framework base on Spark.Through the analysis and preprocessing of the track data,the data is prepared for the subsequent track prediction.Then,according to the characteristics of ship's trajectory,a long short-term memory(LSTM),which can realize long-term memory,is used to construct a trajectory prediction model.(2)In order to extract and learn track characteristics more fully,the WA-LSTM track prediction model was proposed based on LSTM combined with wavelet transform.Firstly,Analyze RNN and LSTM,mainly including RNN model structure,LSTM model structure and training algorithm,LSTM model is chosen as the basic model of track prediction.At the same time,in order to better observe the local characteristics of the trajectory and learn the characteristics of the trajectory more fully,the wavelet transform is added to the LSTM network.Through the characteristics of the multi-resolution analysis of the wavelet transform,the track sequence is decomposed into several subsequences of different frequencies,and the feature of the track sequence is more fully highlighted,and the track prediction is further improved.The WA-LSTM model is proposed in this paper.(3)Concerning the problem of huge sample data and inefficient model training,the distributed thinking is introduced and a method for parallelization of WA-LSTM model based on Spark is proposed.Combining deep learning with distributed memory computing,the storage problem of mass track data is realized through the distributed storage system HBase.And through the data-parallel method based on Spark,the parallel distribution of network training is realized.On the premise of maintaining the prediction accuracy,the training efficiency of the model is greatly improved.(4)Conduct two parts of experiments by using the actual ship AIS information of the Yangtze River J-level waterway from 2016 to 2018 provided by Changjiang Maritime Safety Administration.One part is to verify the prediction accuracy of the WA-LSTM model.Compared with the prediction results of LSTM,RNN,and ARIMA,it is verified that WA-LSTM has higher prediction accuracy,At the same time,it analyzes and compares the results of different numbers of lagging data as inputs.Another part of the experiment is to design two sub experiments to verify that the WA-LSTM model network based on Spark parallelization can effectively improve the training speed without affecting the accuracy of prediction,thus improving the prediction efficiency of the model.