Research on ship track prediction method based on improved PSO-BP algorithm

• Main Authors: Zhu, FaXin, Li, JinYuan, Bi, QinLin, Lai, MinLing
• Format: Conference Proceeding
• Language: English
• Published: SPIE 16.02.2023
• ISBN: 9781510663053; 1510663053
• ISSN: 0277-786X
• DOI: 10.1117/12.2668566

Ship track prediction is an important embodiment of ship intellectualization. Aiming at the problems of slow convergence speed and falling into local minimum of standard particle swarm optimization BP network algorithm, an improved PSO-BP neural network algorithm is proposed. The learning rate in BP network changes with the cosine of weight in PSO algorithm. At the same time, PSO algorithm based on sine function change and adaptive change of learning factor is used to optimize the inertia weight. For the original AIS data, layda criterion and cubic spline interpolation were used to remove the outliers and repair the missing values of AIS data. Experimental verification with AIS data after processing shows that the improved PSO-BP algorithm improves the prediction accuracy, the prediction accuracy of latitude and longitude increases by 15.3% and 17.2% respectively, while reducing the possibility of falling into the local minimum and improving the convergence speed.