Research On Risk Forewarning Of Vessel Collision Based On Random Forest

With the advancement of international trade,the trade between nations becomes prosperous.Water transportation by means of ship are playing an increasingly important role both in international and domestic trade,which becomes an important support of economic development in china.However,while the maritime trade has made great achievements,there are lots of traffic accidents,especially the casualties and economic losses caused by the collision of ships,which have sounded the alarm.The study of risk forewarning system of vessel collision can not only reduce the impact of human factors on collision accidents,but also provide a basis for the decision-making of ship collision avoidance.This paper establishes a risk forewarning model of ship collision based on the ship collision risk index.By using the technology of intelligent identification,the collision risk situation of ships under different situations will be identified and timely forewarning.The research work and innovation of this paper are as follows:(1)According to the complicated nonlinear mapping relationship between vessel collision risk index and its influence factors,the classification and regression tree(CART)algorithm is applied to the prediction of ship collision risk index for the.The fuzzy comprehensive evaluation method is used to evaluate the real collision risk index for the 500 encountering samples collected by this paper,building the collision risk index identification library that contains expert collision avoidance knowledge.Then the CART regression algorithm is used to train the samples in the library,the vessel collision risk index prediction model based on CART will be constructed accordingly.Experimental results shows that compared with GA-BP,GA-SVM collision risk index prediction model,the CART model is better than other models in aspect of prediction accuracy and prediction speed of collision risk index when the feature dimension is low,the sample size is small and the attribute value is missing seriously.(2)In view of the shortcoming of current ship risk forewarning model,the risk forewarning model of vessel collision based on RFFS feature selection algorithm and random forest classifier is studied by this paper.The algorithm of system clustering is applied to the classification of ship collision forewarning level,which breaks the limitation of dividing sample level artificially.The RFFS feature selection algorithm is used to screen the initial forewarning index,which gets a reasonable forewarning index system.On the basis,the random forest classifier is introduced to construct the risk forewarning model of vessel collision.The experimental results show that compared with the traditional BP neural network,SVM,random forest model is superior to other forewarning model in the aspect of prediction accuracy,prediction stability and generalization ability.(3)The optimal anti-collision time is decided based on Collision risk index.The anti-collision decision model is establish combined with mathematical model of steering collision avoidance Finally,an auxiliary collision avoidance decision-making system is established under the MATLAB simulation platform,which is applied for risk forewarning of vessel collision and decision simulation for the collision risk situation of the ship under the situation of head-on,crossing and overtaking.The simulation result verify the validity of the forewarning model of this paper.