针对AUV的6自由度运动非线性和强耦合性，本文提出一种基于蜣螂优化（DBO）的深度双向时域卷积神经网络的非线性系统辨识建模方法。首先，采用双向时域卷积网络（BiTCN）、双向门控循环开关（BiGRU）、注意力机制构建深度双向时域卷积神经网络（Deep-BiTCN），建立AUV6自由度非线性黑箱模型。其次，为提高Deep-BiTCN模型预测的精度，采用DBO算法对模型超参优化。最后，与支持向量机（SVM）和随机森林（RF）模型比较，验证本文运动建模方法的可行性和有效性。实验结果表明，DBO使得Deep-BiTCN算法模型均方根误差（RMSE）、平均绝对百分比误差（SMAPE）降低58.94%、49.22%，决定系数（R2）提高0.73%；本文提出的运动辨识模型精度高、收敛性强，避免运动非线性导致的预报误差大、强耦合运动系统易发散的问题，能为AUV运动提供一种有效的建模与预报方法。

Aiming at the nonlinear and strong coupling characteristics of AUV 6-DOF motion, a Deep bidirectional Temporal Convolutional Networks based on dung beetle optimization (DBO) is proposed in this paper as a nonlinear system identification modeling method. First, a bidirectional Temporal Convolutional Networks (BiTCN), a bidirectional Gated Recurrent Unit (BiGRU), and an Attention mechanism (self-attention) are used to construct a Deep bidirectional Temporal Convolutional Networks (Deep-BiTCN) to establish a 6-DOF nonlinear black-box model of the AUV. Secondly, in order to improve the accuracy of Deep-TCN model prediction, this paper uses DBO algorithm to optimize the model hyper-parameters. Finally, the validity and feasibility of the motion modeling method in this paper are verified by comparing with support vector machine (SVM) and random forest (RF) model. The experimental results show that, compared with Deep-BiTCN, the root mean square error (RMSE) and symmetric mean absolute percentage error (SMAPE) of the DBO-Deep-BiTCN algorithm model are reduced by 58.94% and 49.22%, respectively, and the coefficient of determination (R2) is improved by 0.73%; the AUV 6-DOF motion model based on DBO-Deep-BiTCN has high accuracy and strong convergence, and avoids the motion nonlinearity leads to the problems of large forecast error and easy dispersion of the motion system under strong coupling, which can provide an effective strategy for AUV 6-DOF motion identification.