文献类型：期刊文献

中文题名：基于PSO-RBFNN的船舶横摇运动实时预报

英文题名：Real-time Prediction of Ship Roll Motion Based on PSO-RBFNN

作者：廖声浩[1];王立军[1];王思思[1];贾宝柱[2];尹建川[2];李荣辉[2]

机构：[1]广东海洋大学船舶与海运学院,广东湛江524009;[2]广东海洋大学广东省南海海洋牧场智能装备重点实验室,广东湛江524088

年份：2025

卷号：45

期号：2

起止页码：103

中文期刊名：广东海洋大学学报

外文期刊名：Journal of Guangdong Ocean University

收录：北大核心2023、、北大核心

基金：国家自然科学基金(52171346,52271361);广东省南海海洋牧场智能装备重点实验室资助课题(2023B1212030003);广东省普通高校重点领域项目(2024ZDZX3054)。

语种：中文

中文关键词：船舶横摇运动;运动预报;智能航行;径向基函数神经网络;粒子群优化算法

外文关键词：ship roll motion;motion prediction;intelligent navigation;radial basis function neural network;particle swarm optimization algorithm

中文摘要：【目的】针对船舶横摇运动具有非线性和多变量耦合等特征,提出一种基于粒子群优化(PSO)径向基函数神经网络(RBFNN)的预报模型,以提升预报精度,支持智能航行。【方法】构建基于PSO和RBFNN的混合预报方案。采用PSO对RBFNN的中心和宽度参数进行全局优化,通过PSO-RBFNN模型对船舶横摇运动进行预报。【结果】基于“育鲲”轮实测和仿真数据,验证了模型的可行性和有效性。仿真结果表明,PSO-RBFNN在3种不同工况下均表现出优异的预报性能[提前3 s预报时,平均绝对误差(MAE)≤0.1119,均方误差(MSE)≤0.0280,均方根误差(RMSE)≤0.1673,归一化均方根误差(NRMSE)≤0.0212,平均绝对百分比误差(MAPE)≤22.9%,决定系数(R^(2))≥0.9884],显著优于PSO-RNN、PSO-BP和PSO-MLP等模型。【结论】PSO-RBFNN模型能够高效、准确地预报船舶横摇运动,并在多种工况下保持稳定的性能优势,为智能航行提供实时可靠的技术支撑。

外文摘要：【Objective】To address the nonlinear and multi-variable coupling characteristics of ship roll motion,a real-time prediction model based on particle swarm optimization(PSO)and radial basis function neural network(RBFNN)was proposed to improve prediction accuracy and support intelligent navigation.【Method】This study developed a hybrid prediction scheme based on PSO and RBFNN.PSO was used to globally optimize the center and spread parameters of the RBFNN,and the PSO-RBFNN model was applied to predict ship roll motion.【Result】The feasibility and effectiveness of the proposed model were validated using measured and simulated data from the ship“Yukun”.Simulation results demonstrated that PSO-RBFNN achieved excellent prediction performance[with mean absolute error(MAE)≤0.1119,mean square error(MSE)≤0.0280,root mean square error(RMSE)≤0.1673,normalized root mean square error(NRMSE)≤0.0212,mean absolute percentage error(MAPE)≤22.9%and coefficient of determination(R^(2))≥0.9884 for a 3-second ahead forecast]under three different conditions,significantly outperforming models such as PSO-RNN,PSO-BP,and PSO-MLP.【Conclusion】The PSO-RBFNN model can efficiently and accurately predict ship roll motion,maintain stable performance under various operating conditions.It provides real-time and reliable technical support for intelligent navigation.