Research On Collision Prediction Methods For Inland Vessels Based On AIS Data

With the accelerated development of intelligent inland waterway vessels,certain achievements have been made in areas such as inland environmental perception,vessel trajectory prediction,and collision risk prediction.Utilizing multi-sensor fusion or AIS(Automatic Identification System)in combination with trajectory prediction to acquire both dynamic and static information about vessel navigation can further enhance the safety of inland waterway navigation and relieve the pressure on crew members.While multi-sensor perception schemes can perceive nearby vessel navigation information in real time,the high cost hinders its widespread application in inland waters.Conversely,AIS is relatively cost-effective and has become ubiquitous among inland waterway vessels.Therefore,researching collision prediction methods for inland vessels based on AIS data has substantial practical value.The distance to the closest point of approach(DCPA)and the time to the closest point of approach(TCPA)are important indicators for assessing the potential collision risk between two vessels.They denote the distance and time to the closest point of approach(CPA),respectively,and can be used for collision prediction.However,current scholars calculate DCPA and TCPA parameters through the vessel’s geographical coordinates,overlooking the error brought by the vessel’s size.In narrow inland waterways,conventional DCPA and TCPA collision detection methods can’t provide auxiliary support for the safe navigation of water vessels.To address the above issues,the main research content of this article is as follows:(1)Due to limited data collection and sharing channels leading to insufficient vessel navigation data resources,a vessel navigation data collection system based on onboard AIS equipment has been designed.Installed on inland transport vessels,it collects both static and dynamic navigation information about the vessel and nearby ones.The received message information is acquired through serial communication,filtered,parsed,and stored,then sent remotely through cellular communication and intranet penetration technologies.The collected data is used for experimental verification of inland vessel collision prediction methods.(2)To obtain real-time collision prediction with the vessel and the other vessels at the same moment,and to improve the accuracy of vessel position estimation,an extended Kalman filter-based navigation position prediction method has been researched.It uses a vessel motion model for state estimation and AIS data as observation data to correct estimation errors.Experimental results show that,in the absence of AIS data input,the average position error of the state estimation is 6.679 m,while it is 1.865 m when AIS data is used as observation data.(3)To address the issue of poor accuracy in narrow waters due to traditional CPA parameter calculations not considering vessel size information,a new CPA parameter calculation method is proposed.It corrects CPA parameters by calculating the minimum distance between vessel outlines.With an average single-use time under 1.74 s in different scenarios,it has good real-time performance.The corrected parameters can assist in predicting collisions of inland vessels and can also be used for the relevant models of the vessel risk prediction index method to enhance the accuracy of collision risk assessment.