| 引用本文: | 姜述强,金鸿章,魏凤梅.可容错的遥控水下机器人递归神经网络控制[J].哈尔滨工业大学学报,2013,45(9):57.DOI:10.11918/j.issn.0367-6234.2013.09.011 |
| JIANG Shuqiang,JIN Hongzhang,WEI Fengmei.A fault-tolerable recurrent neural network controller for remote operated vehicle[J].Journal of Harbin Institute of Technology,2013,45(9):57.DOI:10.11918/j.issn.0367-6234.2013.09.011 |
|
| 摘要: |
| 针对遥控水下机器人(ROV)需要长时间稳定可靠工作的问题,提出递归模糊神经网络及可容错分配推力的控制方法．使用扩展函数链改进递归模糊神经网络控制器,提高了控制器对机器人非线性特性的识别和处理能力;基于反向梯度传播原理,由能量函数设计了该网络的学习算法,并根据微粒群优化确定学习率参数,从而保证整个网络的收敛性;在推力分配方面,针对开架式遥控水下机器人的两种推力器布置形式进行建模,将容错问题转化为对偶优化问题,建立能量函数实现故障条件下的推力优化分配．实验结果表明,所设计控制器不仅增强了遥控水下机器人对干扰的反应能力,并且提高了对机器人非线性特性的控制能力,减少了控制误差．当部分主推或侧推等推力器失效时,仍可以通过推力优化分配实现机器人在水平面上的准确位置控制,从而保证了遥控水下机器人长时间可靠工作． |
| 关键词: 遥控水下机器人 递归神经网络 扩展函数链 推力分配 容错控制 |
| DOI：10.11918/j.issn.0367-6234.2013.09.011 |
| 分类号: |
| 基金项目:国家自然科学基金资助项目(9,0);高等学校博士学科点专项科研基金(20122304120003). |
|
| A fault-tolerable recurrent neural network controller for remote operated vehicle |
|
JIANG Shuqiang1, JIN Hongzhang1, WEI Fengmei2,3
|
|
(1. College of Automation,Harbin Engineering University, 150001 Harbin, China; 2. College of Computer Science and Technology, Harbin Engineering University, 150001 Harbin, China; 3.School of Technology,Harbin University, 150086 Harbin, China)
|
| Abstract: |
| A fuzzy recurrent neural network controller with fault tolerable thrust allocation control strategy has been proposed for Remote Operated Vehicles (ROV). Extended functional link is issued for fuzzy recurrent neural network controller to improve the ability of identification and response. Online training algorithm is developed based on gradient descent method. The learning rate parameters are determined according to particle swarm optimization, hence the whole network convergence is guaranteed. On the aspect of force allocation, a model has been established according to thruster positions for open frame remote operated vehicles based on which the fault tolerable problem has been transformed into dual optimization problem. The energy function concerning ROV control has been established so as to obtain thrust optimal allocation under fault conditions. Experiments have demonstrated that the controller can improve the ROV capacity to handle nonlinear characteristics and strong disturbance, reduce control errors. When one of main or side thrust fails, the controller can still accomplish precise horizontal position control through thrust optimization allocation strategy. Therefore, the reliable operation of ROV for the long time-span has been improved. |
| Key words: ROV recurrent neural network Extended functional link duality principle thrust allocation fault tolerance control
|
