基于灰狼优化支持向量回归的船舶航迹预测
2021-01-06陈影玉索永峰杨神化
陈影玉 索永峰 杨神化








摘要:为提高船舶航迹预测精度,利用灰狼优化(grey wolf optimization, GWO)算法求出支持向量回归(support vector regression, SVR)模型的最优参数,构建基于GWO-SVR的船舶航迹预测模型。选取福建漳州古雷港水域某船航迹的AIS数据。将该模型与其他模型的预测结果进行比较。实验结果表明,与SVR、灰狼优化最小二乘支持向量机和粒子群优化最小二乘支持向量机等3种模型相比,GWO-SVR模型的预测精度分别提升2.61%、10.93%和0.22%,预测误差分别降低0.022、0.303和0.172。本文方法提高了船舶航迹预测精度,可为海事监管人员及时作出正确决策和保障航行安全提供支持。
关键词: 船舶航迹预测; 支持向量回归(SVR); 灰狼优化(GWO)
中图分类号: U676.1 文献标志码: A
Abstract: In order to improve the accuracy of ship trajectory prediction, the optimal parameters of the support vector regression (SVR) model are found by the grey wolf optimization (GWO) algorithm, and a ship trajectory prediction model based on GWO-SVR is constructed. The AIS data of trajectory of a ship of Gulei Port in Zhangzhou of Fujian province are selected.The model prediction result is compared with prediction results by other models. Experimental results show that, the prediction accuracy of GWO-SVR model is improved by 2.61%, 10.93% and 0.22%, and the prediction error is reduced by 0.022, 0.303 and 0.172, respectively, compared with SVR, the GWO-least squares support vector machine and the particle swarm optimization least squares support vector machine. The method in the paper improves the accuracy of ship trajectory prediction and can provide support for maritime supervisors to make correct decisions and ensure navigation safety in time.
Key words: ship trajectory prediction; support vector regression (SVR); grey wolf optimization (GWO)
0 引 言
為了将无人船引入商业航线,必须建立有效的避碰系统,以确保无人船自主作业所需的安全水平[1]。船舶航迹预测结果可以用来评估船舶碰撞风险,支持主动避碰,增强海上情境态势感知的能力,为船舶避碰决策提供参考。
目前船舶航迹预测方法可分为3类:基于统计模型、基于概率图模型和基于机器学习模型的预测方法。基于统计模型的预测方法有高斯回归[2]、卡尔曼滤波[3-4]和贝叶斯网络[5]等,其特点为:在建模时一般需要考虑风、流等环境对船舶运动的影响;船舶运动学方程的建立会增加模型的复杂度;随着实验的进行,误差也会不断增加。基于概率图模型的预测方法有马尔科夫模型[6]和灰色模型[7]等,其特点为:它结合概率论与图论,将不同领域的知识抽象为概率模型,将应用中的问题归结为计算概率模型变量的概率分布;灰色模型在数据量大的情况下性能较低,运行时间较长,但改进后的预测效果优于最小二乘法的预测效果。……
