基于SCI-CA模型的船舶纵摇多维多步预测方法  

Multi-dimensional and multi-step prediction method for ship pitching based on SCI-CA model

作　　者：王宇超[1] 赵洵 杨周琦 傅荟璇[1] WANG Yu-chao;ZHAO Xun;YANG Zhou-qi;FU Hui-xuan(College of Intelligent Systems Science and Engineering,Harbin Engineering University,Harbin 150000,China)

机构地区：[1]哈尔滨工程大学智能科学与工程学院,哈尔滨150000

出　　处：《控制与决策》2025年第1期64-70,共7页Control and Decision

基　　金：国家自然科学基金项目(52271313);中央高校基本科研业务费专项资金项目(3072024GH0405)。

摘　　要：海洋环境复杂多变,船舶航行容易受到风浪、洋流等因素的干扰,船舶运动具有非线性、耦合性等特点.针对传统的船舶运动姿态预测方法对时序数据的提取效率尚有不足,难以达到高精度预测效果的问题,提出样本卷积交互-通道注意力(SCI-CA)神经网络船舶纵摇运动预测模型.该模型采用多类别船舶运动姿态数据作为输入,将输入拆分为两个子序列,利用样本卷积交互网络(SCI)的递归下采样卷积交互结构,结合多分辨率聚合而成的丰富特征,提高船舶运动数据深层特征的利用率.再通过通道注意力机制(CA)提高有效通道的权重比例,并以残差结构输入到全连接层,得到最后的预测结果.实船数据验证结果表明,SCI-CA组合模型预测结果较其他模型预测精度高,其平均绝对百分比误差(MAPE)、均方根误差(RMSE)均有明显降低,验证了SCI-CA模型预测船舶运动的有效性.As the marine environment is complex and changeable,ship navigation is easily affected by factors such as wind,waves,ocean currents and other factors,and ship motion is characterized by nonlinearity and coupling.Aiming at the problem that traditional ship motion prediction methods have insufficient efficiency in extracting time series data and are difficult to achieve high-precision prediction results,a sample convolution and interaction-channel attention(SCI-CA)neural network ship pitch motion prediction model is proposed.The model uses multi-category ship motion attitude data as input,splits the input into two subsequences,utilizes the recursive down sampling convolution interaction structure of the sample convolution interaction network(SCI),and combines the rich features aggregated from multiple resolutions to improve the utilization of deep features of ship motion data.Then,channel attention mechanism(CA)is used to improve the weight ratio of effective channels,and the residual structure is input to the full connection layer to obtain the final prediction result.The simulation results of real ship data show that the prediction accuracy of the SCI-CA combined model is higher than that of other models,and its mean absolute percentage error(MAPE)and root mean square error(RMSE)are significantly reduced,verifying the effectiveness of the SCI-CA model in predicting ship motion.

关 键 词：船舶纵摇 SCI-Net 通道注意力 交互学习结构 组合模型 多步预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]