Incremental echo state network for a water-jet propulsion USV: Theoretical and experimental research | |
Li DC(李德才); He YQ(何玉庆) | |
2016 | |
会议名称 | 2016 American Control Conference, ACC 2016 |
会议日期 | July 6-8, 2016 |
会议地点 | Boston, MA, United states |
页码 | 5296-5301 |
通讯作者 | 李德才 |
中文摘要 | In this paper, a new system identification method is proposed. Since the new method combines the advantage of Kalman filter, l1 norm regularization and echo state mechanisms, it termed as ESNKF-l1. The ESNKF-l1 inherits the basic idea of ESN learning with Kalman filter to update model parameters online and utilize l1 norm regularization to obtain a sparse model structure. Moreover, the training of the ESNKF-l1 is facilitated by employing a bound optimization algorithm, based on which, a proper surrogate function is defined and the l1 norm regularization is approximated by l2 norm, while remaining constrains to the model parameters. It lead to an efficient method for model parameters estimation, which can be solved by using pseudo-observation (PO) method. In this case, the l2 norm is incorporated into the filtering process by designing a fictitious measurement function and the model parameters can be updated in an iteration manner. Experimental results show that the proposed method is suitable to capture the dynamics of USV and is superior to existing methods. |
收录类别 | EI ; CPCI(ISTP) |
产权排序 | 1 |
会议主办者 | Adaptics; et al.; GE Global Research; MathWorks; Mitsubishi Electric Research Laboratory (MERL); Quanser |
会议录 | Proceedings of the American Control Conference |
会议录出版者 | IEEE |
会议录出版地 | NEW YORK |
语种 | 英语 |
ISSN号 | 0743-1619 |
ISBN号 | 978-1-4673-8682-1 |
WOS记录号 | WOS:000388376105059 |
内容类型 | 会议论文 |
源URL | [http://ir.sia.cn/handle/173321/19364] |
专题 | 沈阳自动化研究所_机器人学研究室 |
推荐引用方式 GB/T 7714 | Li DC,He YQ. Incremental echo state network for a water-jet propulsion USV: Theoretical and experimental research[C]. 见:2016 American Control Conference, ACC 2016. Boston, MA, United states. July 6-8, 2016. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论