Motion control of underwater vehicle based on least disturbance BP algorithm

Abstract

Up to now, some technology of neural networks are developed to solve the non-linearity of researched objects and to implement the adaptive control in many engineering fields, and some good results were achieved. Though it puts some questions over to design application structure with neural networks, it is really unknowable about the study mechanism of those. But, the importance of study ratio is widely realized by many scientists now, and some methods on the modification of that are provided. The main subject is how to improve the stability and how to increase the convergent rate of networks by defining a good form of the study ratio. Here a new algorithm named LDBP (least disturbance BP algorithm) is proposed to calculate the ratio online according to the output errors, the weights of network and the input values. The algorithm is applied to the control of an autonomous underwater vehicle designed by HEU. The experimental results show that the algorithm has good performance and the controller designed based on it is fine.

Keywords

BP algorithm of neural networks / dynamic ratio / least disturbance / autonomous underwater vehicle

References

[1] Zhang Nai-yao, Yan Ping-dan. Neural networks and fuzzy control[M]. 1998, Beijing: Tsinghua Press

[2] Peng Liang, Lu Ying-chun, Wan Lei. Control with neural network for autonomous underwater vehicle [J]. Marine Engineering, 1995, 13(2): 38-46

[3] Fujii T, Ura T. Control with neural network for autonomous underwater vehicle [J]. J Soc Naval Arch, 1989, 166: 503-511

[4] Yuh J. A neural net controller for underwater robotic vehicles[J]. IEEE J Oceanic Eng, 1994, 15(3): 161-166

[5] LIU Xue-min, WAN Lei, XU Yu-ru. Application of Fuzzy neural network in the design of autonomous underwater vehicles’ controller[J]. Central South Industry University Transactions, 2000(3): 182–184.

[6] Noriega J R, Wang H. A direct adaptive neuralnet-work control for unknown nonlinear systems and its application[J]. IEEE Transactions on Neural Networls, 1998, 9(1): 27-34

[7] Ali Ö G, Chen Y T. Design quality and robustness with neural networks[J]. IEEE Transactions on Neural Networks, 1999, 10(6): 1518-1527

[8] Song Q, Xiao J Z, Soh Y C. Robust backpropagation training algorithm for multilayered neural tracking controller[J]. IEEE Transaction on Neural Networks, 1999, 10(5): 1133-1141

[9] Hou C H, Qian J X. Stability analysis for neural dynamics with time-varying delays[J]. IEEE Transactions on Neural Networks, 1998, 9(1): 221-223