计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (6): 258-263.DOI: 10.3778/j.issn.1002-8331.1607-0268

• 工程与应用 • 上一篇    下一篇

AUV深度的神经网络辨识和学习控制仿真研究

曾德伟,吴玉香,王  聪   

  1. 华南理工大学 自动化科学与工程学院,广州 510640
  • 出版日期:2017-03-15 发布日期:2017-05-11

Research on simulation of neural network identification and learning for AUV depth control

ZENG Dewei, WU Yuxiang, WANG Cong   

  1. College of Automation Science and Technology, South China University of Technology, Guangzhou 510640, China
  • Online:2017-03-15 Published:2017-05-11

摘要: 将自主水下航行器(AUV)的深度控制问题转换为对非线性严格反馈系统的分析,提出了一种结合反步法和确定学习理论的自适应学习控制方法。通过反步法设计了一种输入状态稳定(ISS)神经网络控制器,其中引入小增益定理,避免了控制器设计中存在的奇异值问题,并在满足持续激励(PE)条件下,利用神经网络辨识实现了对系统未知动态的局部准确逼近和部分神经网络权值的收敛,保证了闭环系统的稳定。将从动态模式中学到的知识静态保存,提取动态特征设计学习控制器,仿真结果表明,该控制器避免了执行同样任务时的重复训练,改善了系统控制性能,验证了所提控制方法的有效性。

关键词: 自主水下航行器, 反步法, 确定学习, 神经网络辨识, 学习控制

Abstract: Converting the depth control problem of Autonomous Underwater Vehicles(AUV) to the analysis of nonlinear strict-feedback system, an adaptive learning control method based on backstepping and deterministic learning is proposed. The Input-to-State Stability(ISS) modular neural network controller that is designed through backstepping method achieves the convergence of partial neural weights and locally-accurate approximation of unknown system dynamics by using neural network identification under the Persistent Excitation(PE) condition, and in order to avoid the controller singularity problem, the ISS-type small gain theorem is used to guarantee the stability of the closed-loop system. The learned knowledge acquired from the dynamical pattern can store as constant neural weights, which can be used to design learning controller. The simulation results show that the controller avoids duplication of training to perform the same task, which improves the performance of the system and verifies the effectiveness of the method.

Key words: autonomous underwater vehicles, backstepping, deterministic learning, neural network identification, learning control