Study On Evaluation For Severity Level Of Maritime Traffic Flow Conflict Based On BP Neural Network

Recently,with the boost of China economics and applying the nation strategy of strengthening the ocean and shipping,it requires a higher level of maritime traffic safety.The increasing complexity of traffic flow on the sea,however,meets a more difficult maritime traffic safety management since the modern vessels trending to high speed and large-scale.In that way,study on the risk of maritime traffic calls for a perspective method to be better at avoiding maritime traffic accidents and improving the navigation safety on the seas.Assessing the severity of maritime traffic flow conflict is the foundation work of revealing the characteristics,regularity,mechanism and evolution of maritime traffic flow.It can not only analyze the maritime traffic safety state in the seas providing the decision-making support,but also make positive reference effects on preventing maritime traffic accidents.The existing maritime traffic safety assessing method is difficult to apply when the relationship among evaluating indices is vague or the source is not clear,and the high subjectivity may deviate the assessment results.In that way,this thesis made an analysis to the influence factors of maritime traffic flow conflict,extracting 11 factors like vessels type distribution,dimension distribution,velocity distribution,course distribution and so on.Finally,it established an assessing model of the severity of maritime traffic flow conflict basing on BP neural network.Considering that the big diversity among different seas,during BP neural network training,the difference of original data would affect the training efficiency and assessing accuracy.So,in this thesis the training data were classified by Euclidean Metric to lessen the diversity of different data then the classified data would be trained and verified.In order to assess the severity of maritime traffic flow conflict comprehensively,many influence factors were extracted,but this would increase the complexity of neural network calculation since there are too many factors on input level.In that way,the PCA(principal component analysis)was applied here to make a dimensionality reduction to the influence factors,and some principal factors would represent the original ones.After classifying,dimensionality reduction,and function comparing through network training,it met the requirement of function accuracy and iterations so that the final assessment model was established.In this thesis,the historical statistics data from 9 seas were applied to verify.From the comparison of evaluating accuracy in BP neural network between original data and cluster dimensionality reduction respectively,it showed that the average error of evaluating accuracy was decreased from 10.9% to 2.8% and it indicated that the applying of BP neural network is verifiable.Furthermore,the results showed that this model,combining cluster analysis,PCA and BP neural network,is not only more objective,but also able to increase the accuracy of assessment rather than the single BP neural network model.At last,this model was used to assess and analyze the data from the waterway of Lao Tieshan to Jingtang Port in 2016.