Research On Real-time Detection And Prediction Methods Of Intelligent Ship Collision Ris

The maturity of the new generation of artificial intelligence technology has promoted the rapid development of maritime transportation in the direction of intelligence.As a carrier of maritime transportation,intelligent ships have important advantages such as reducing labor costs,energy saving and emission reduction,and their transportation safety issues have also become a research topic of general concern at home and abroad.Among various safety issues,ship collision accidents occur most frequently and cause immeasurable losses.Due to the complexity and diversity of ship encounters,the research on automatic collision avoidance in the navigation process of intelligent ships is one of the technological difficulties.The collision avoidance system is mainly divided into two stages,collision risk detection and collision avoidance decision-making.This research is dedicated to solving the problem of perception of collision risk of intelligent ships.Based on the automatic identification system（AIS）,a real-time detection and prediction model for collision risk of intelligent ships during navigation is established.The main research contents and innovations of this paper are as follows:1)Most of the current researches on ship domains use the boundary of the ship domain as the critical value of collision risk.This type of ship domain usually uses a model to solve all encounter scenarios,which does not match the complicated sailing situation.This chapter proposes a dynamic ship domain concept based on risk perception and an analysis framework under different encounter situations.Based on the Cloeset Point of Approach（CPA）detection of potential conflict accidents,the ship’s avoidance maneuvering space is estimated,and different encounter situations are classified according to the International Regulations for Preventing Collisions at Sea（COLREGs）.The statistical analysis results explore the dynamic boundaries of the ship domain in different situations.This risk perception-based ship domain can support conflict detection and conflict resolution,providing new insights for navigators’behavioral decisions in collision avoidance environments.2)The proposed risk assessment models based on AIS data are mostly used to analyze high-risk areas in a given sea area.In the actual navigation of an intelligent ship,in addition to the understanding of the sea area situation,the detection of the possibility of real-time collision when the encountering ship is approaching is really important.With the reliable source of AIS data and lower latency,real-time collision risk detection becomes possible.Therefore,based on the traditional risk assessment algorithm,we propose an exponential factor about the safety distance.The model uses the K-MEANS algorithm to cluster the sailing speed and classifies six different sailing conditions.This factor is combined with a data-driven risk quantification algorithm to detect collision probability in real time.By analyzing a large amount of historical AIS data to classify the sea area conflict risk level（CRL）,so as to provide a basis for CRL judgment under actual navigation conditions.3)Due to the large volume and mass of ships,there is a large delay in steering or braking.The untimely collision avoidance measures have led to most collisions.In order to further ensure the navigation of intelligent ships,an algorithm for predicting the collision risk of encountering ships is proposed.Based on the historical navigation data of paired ships（including related navigation data such as latitude and longitude,speed and course angle）,combined with the extended Kalman filter algorithm（EKF）based on polynomial fitting,it can perform the optimal prediction of the navigation data at next moment(T_k+1).The predicted result will be used as the input of the real-time collision risk detection model（Rt CR）for collision risk detection at T_k+1.Based on the historical navigation risks of the encountering ships,the T_k+2collision risk prediction is further combined with the unscented kalman filter（UKF）algorithm.As a result,the model has been combined with the filtering algorithm twice to complete the two-step prediction of the collision risk of the encountering ship under the premise of ensuring the accuracy of the prediction.