| 罗文龙,章政,付永鹏,张军生,刘思贤.基于NARX动态神经网络的球形机器人混合建模[J].高技术通讯(中文),2026,36(6):643~651 |
| 基于NARX动态神经网络的球形机器人混合建模 |
| Hybrid modeling of spherical robot based on NARX dynamic neural network |
| |
| DOI：10. 3772 / j. issn. 1002 - 0470. 2026. 06. 010 |
| 中文关键词: 球形机器人； 自适应小波包去噪； 动态神经网络； 混合建模 |
| 英文关键词: spherical robot, adaptive wavelet packet denoising, dynamic neural network, hybrid modeling |
| 基金项目: |
| 作者 | 单位 | | 罗文龙 | （武汉科技大学信息科学与工程学院武汉 430081） | | 章政 | | | 付永鹏 | | | 张军生 | | | 刘思贤 | |
|
| 摘要点击次数: 78 |
| 全文下载次数: 71 |
| 中文摘要: |
| 球形机器人具有欠驱动、非完整约束、数学模型复杂等特点，针对球形机器人存在难以建立精确数学模型的问题，提出了一种基于机理建模和外因输入的非线性自回归(nonlinear auto-regressive with exogeneous inputs，NARX) 动态神经网络数据驱动建模的混合建模方法。首先，基于所搭建的球形机器人平台，使用拉格朗日方程建立其动力学模型；然后，针对机理建模过程中存在机理假设和非线性环节处理导致建模偏差，设计了一种自适应小波包去噪（adaptive wavelet packet denoising，AWPD）算法和NARX动态神经网络的数据驱动建模方法，通过采用自适应阈值函数策略，能够根据每一层的噪声分布规律自适应地调整AWPD算法的阈值，提升数据的去噪效果，同时，采用NARX动态神经网络建立了球形机器人的数据驱动模型，在此基础上将动力学模型和数据驱动模型相结合构建球形机器人混合模型。实验结果表明，相比于传统的机理建模和数据驱动建模方法，本文所设计的混合建模方法具有更强的数据去噪能力、更高的建模精度和泛化能力。 |
| 英文摘要: |
| Spherical robot has the characteristics of underactuation, non-holonomic constraint, complex mathematical model, etc. In order to solve the problem that it is difficult to establish accurate mathematical model, a hybrid modeling method based on nonlinear auto-regressive with exogeneous inputs (NARX) dynamic neural network data-driven modeling is proposed. Firstly, based on the spherical robot platform, Lagrange equation is used to establish its dynamics model. Then, aiming at the modeling deviation caused by mechanism assumption and nonlinear processing, a adaptive wavelet packet denoising (AWPD) algorithm and a data-driven modeling method based on NARX dynamic neural network are designed. By adopting the adaptive threshold function strategy, the threshold value of AWPD algorithm can be adjusted adaptically according to the noise distribution law of each layer to improve the noise reduction effect of data. Meanwhile, the data-driven model of spherical robot is established by using NARX dynamic neural network. On this basis, the dynamic model and data-driven model are combined to build a hybrid model of spherical robot. Finally, the experimental results show that the hybrid modeling method designed in this paper has stronger data denoising ability, higher modeling accuracy and generalization ability compared with the traditional mechanism modeling and data-driven modeling methods. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
