基于滑动窗口PSO-LSSVR的船舶轨迹预测模型  

Ship Trajectory Prediction Model Based on Sliding Window PSO-LSSVR

作　　者：王均刚 丁惠倩 胡柏青[1] WANG Jun-gang;DING Hui-qian;HU Bai-qing(School of Electrical Engineering,Naval University of Engineering,Wuhan 430033,China;School of Intelligent Science and Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]海军工程大学电气工程学院,武汉430033 [2]哈尔滨工程大学智能科学与工程学院,哈尔滨150001

出　　处：《武汉理工大学学报》2022年第12期35-43,59,共10页Journal of Wuhan University of Technology

摘　　要：传统的船舶轨迹预测模型难以确定关键参数,且训练时间较长,训练样本需求量大。例如神经网络模型,需要大量数据进行训练。为解决这些问题,提高目标船只的预测精度,提出了一种基于粒子群优化(Particle Swarm Optimization, PSO)的最小二乘支持向量回归(Least Square Support Vector Regression, LSSVR)预测模型。提取船舶自动识别系统(Automatic Identification System, AIS)数据中的经度、纬度、航速、航向作为特征量,利用AIS的时间戳作为数据处理的依据进行数据预处理。利用PSO算法对模型进行优化,确定最优参数,采用滑动窗口输入输出的方式进行预测。选取某船舶实际航行的AIS数据进行实验,并与粒子群优化的长短时记忆神经网络(Long Short-Term Memory, PSO-LSTM)模型、PSO-LSSVR模型得到的结果进行对比,结果表明滑动窗口PSO-LSSVR模型预测精度更高,且用时更短。The traditional ship trajectory prediction model is difficult to determine the key parameters.And the training time is long while the demand for training samples is large.For example, neural network models require a lot of data for training.To solve these problems and improve the prediction accuracy of target ships, a least squares support vector regression(LSSVR) prediction model based on particle swarm optimization(PSO) is proposed.The longitude, latitude, speed, and heading in the Ship Automatic Identification System(AIS) data are extracted as feature quantities And the AIS timestamp is used as the basis for data processing for data preprocessing.The PSO algorithm is used to optimize the model to determine the optimal parameters Use the sliding window input and output method to predict the location.The AIS data of the real sailing of a certain ship is selected for experiments.Comparing with the LSTM model and the LSSVR model, the results show that the sliding window PSO-LSSVR model has higher prediction accuracy and shorter time.

关 键 词：滑动窗口 粒子群优化算法 最小二乘支持向量回归 轨迹预测 

分 类 号：U675.96[交通运输工程—船舶及航道工程]