Research On Port Vessel Flow Prediction Based On Deep Learning

In recent years,with the rapid development of economy,the scale of marine traffic continues to expand,and more and more vessels enter and leave the port,causing traffic congestion and even accidents near the port.Accurate vessel flow prediction is an important part of maritime intelligent transportation system,which can provide effective basis for maritime supervision and plays a vital role in the efficiency and safety of maritime transportation.Most of the existing methods of vessel flow prediction are based on AIS data.However,these methods only consider the temporal correlation of vessel flow,ignoring the spatial correlation of vessel flow,which leads to poor prediction performance.In order to solve the above problems,this thesis uses deep learning method to realize the short-term prediction of vessel flow,and designs a vessel flow prediction software.The specific research works are as follows:Firstly,according to the characteristics of temporal and spatial correlation of vessel flow,this thesis designs a model of vessel flow prediction based on ConvLSTM.This thesis preprocesses the AIS data in the prediction area based on grid division to obtain vessel flow data.Then the ConvLSTM effectively extracts the temporal and spatial characteristics of vessel flow to implement the multi-step prediction of vessel flow.The experimental results show that the model based on ConvLSTM has achieved a good performance.Secondly,in order to improve the prediction performance,based on the ConvLSTM prediction model,this thesis proposes a model of vessel flow prediction based on attention mechanism.The weight of different time steps is obtained by attention mechanism to solve the problem of prediction performance degradation caused by long input sequence.At the same time,this thesis optimized the model by adjusting the parameters.The experimental results show that,compared with the prediction model based on ConvLSTM,the proposed ConvLSTM prediction model based on attention mechanism has better prediction performance,and its evaluation metrics RMSE and MAE reach 1.4605 and 0.9303 respectively,which verify the effectiveness of attention mechanism.Compared with the existing prediction models,the proposed model has higher prediction accuracy.Finally,according to the actual demand,this thesis designs and implements a vessel flow prediction software.The software uses Django framework to implement the back-end,MySQL as the database,and Bootstrap framework to develop the web interface,which can provide users with vessel flow prediction,historical flow query,user management and other functions.The system test verifies the effectiveness of the designed software.In this thesis,deep learning technology is used to implement the vessel flow prediction method,which has high prediction accuracy.The designed vessel flow prediction software is helpful to improve the efficiency and safety of maritime traffic management.