| 神经网络在水下机器人控制中的应用研究 | |
| Alternative Title | Study on AppIication of Neural Networks in Control of Autonomous Underwater Vehicles |
| 叶志超1,2 | |
| Department | 水下机器人技术研究室 |
| Thesis Advisor | 封锡盛 |
| Classification | TP242.3 |
| Keyword | 自治水下机器人 神经网络 内模控制 |
| Call Number | TP242.3/Y41/2001 |
| Pages | 62页 |
| Degree Discipline | 模式识别与智能控制 |
| Degree Name | 硕士 |
| 2001 | |
| Degree Grantor | 中国科学院沈阳自动化研究所 |
| Place of Conferral | 沈阳 |
| Abstract | 水下机器人系统是非线性和时变的动态系统，各自由度之间存在耦合，在运行时常受到外界环境的干扰。因此水下机器人的控制是一个很复杂的问题。作者在论文工作进行的过程中，首先针对“CR－01”6000水下机器人详细地研究了其航行控制系统，对“CR－01”控制系统软件，主要是自动驾驶系统、机载设备控制系统的软件进行了剖析。深入系统地学习了神经网络控制方面的基本理论，以“CR－01”的数学模型为受控对象进行了神经网络内模控制的仿真研究，取得了较好的仿真结果。通过仿真结果说明了神经网络内模控制方案对“CR－01”模型的有效性，同时也说明了构成控制系统的神经网络结构和训练神经网络时训练样本生成策略的有效性。在提高神经网络权值训练速度方面进行了一些探索，将BFGS变尺度法应用到神经网络权值训练上，和普通的误差反传算法相比，训练速度有明显的提高。论文的实验工作是在远程AUV上进行的。在实验之前，先对远程AUV的实验数据进行了辨识，辨识结果表明，用和“CR－01”数学模型相同结构的数学模型（即三阶模型）来对远程AUV的实验数据进行辨识可以获得比较高的精度。同时，用六阶模型进行辨识的结果与三阶模型相比，精度并没有明显的提高。因此，在对“CR－01”进行仿真中所用的神经网络结构可以直接用来训练远程AUV的正模型和逆模型，进而构成远程AUV的航向回路控制系统。在利用已获得的实验数据对神经网络重新训练后构成内模控制系统对远程AUV进行了水池实验，实验结果验证了神经网络内模控制方案的有效性。 |
| Other Abstract | Control problems of underwater vehicles have difficulties dew to the non-linearity and time-variety of dynamics property of underwater vehicles. Besides, the coupling between different degrees and the disturbance from the environment make the design of control system a challenging task. During working on the dissertation, the author first studied the heading control system of "CR-01", an AUV, and the theory of neural network control. In the dissertation, the author presents the simulation of neural network internal model control (IMC) system that took the mathematic model of "CR-01" as controlled object. The result of simulation shows the effectivity of the structure of neural networks composing the control system and the effectivity of the scheme choosing the sample for training the weights of neural networks. To improve the convergence speed of training, the author applied the BFGS variable metric method when training neural networks and got obvious enhance on convergence speed compared with normal back-propagation method. The experiment is completed with the Long Distance AUV. Before experiment identification of previous data was executed and the result showed that a high precision could be obtained when identifying the dynamics of Long Distance AUV with the same mathematic model (3-order) as that of "CR-01". And identification showed that there is not an evident difference between the precision of 6-order model and that of 3-order model. So the structure used in the simulation of "CR-01" can be trained as the model or anti-model of Long Distance AUV and then form the control system of heading control loop.After being trained with previous data again, neural networks formed the internal model control system of Long Distance AUV whose effectivity was validated in experiment executed in the pool lab. |
| Language | 中文 |
| Contribution Rank | 1 |
| Document Type | 学位论文 |
| Identifier | http://ir.sia.cn/handle/173321/644 |
| Collection | 水下机器人研究室 |
| Affiliation | 1.中国科学院沈阳自动化研究所 2.中国科学院研究生院 |
| Recommended Citation GB/T 7714 | 叶志超. 神经网络在水下机器人控制中的应用研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2001. |
| Files in This Item: | ||||||
| File Name/Size | DocType | Version | Access | License | ||
| 神经网络在水下机器人控制中的应用研究.p（2884KB） | 开放获取 | -- | Application Full Text | |||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment