Research On Intelligent Vessel Collision Avoidance Algorithm Based On AIS Ship Navigation States

Although rapidly developing shipping industry has adapted to international trade,it has caused great risks to maritime traffic safety,so the research on ship collision avoidance has been an important research topic.Collision avoidance decision-making technology is a kind of ship autonomous navigation technology system.In order to make the collision avoidance decision system run faster and have better knowledge expression effect,intelligent algorithm and traditional algorithm are often combined to study ship collision avoidance.The technical difficulty of collision decision is the choice of collision avoidance time which is determined by collision risk index.Based on the AIS data of real collision accidents.Three comparative experiments are designed to improve the accuracy of the training model.The advantage of LSTM neural network in predicting time series is used to predict ship collision risk.The intermediate data required for calculating the collision risk is used as the network input.The experimental results are smaller than the mean square error of AIS initial data trained by BP neural network;Training AIS initial data suitable for open waters and the initial data suitable for restricted waters and considering the ship-ship interaction by BP neural network.The results show that the MSE of the latter is smaller;The LSTM neural network is proposed to predict the future navigation trajectory of the two ships,and the collision risk model is used to learn the prediction data.The final MSE of this experiment is smaller than that of the neural network trained directly with AIS data.Finally,the collision avoidance decision-making program is used to verify the collision avoidance effect of the real collision accidents in the three encountering situations : encountering,overtaking and crossing situation.The results show that the effect of predicting the future trajectory first and then calculating the collision risk index can achieve the dynamic prediction of the collision risk of the two ships in advance.