Abstract：Considering the highly nonlinear and time-variant dynamics of autonomous underwater vehicles(AUVs),a generalized predictive control strategy based on neural network is presented.The modified Elman neural network is used as the multi-step predictive model and the neural network learning algorithm is improved.Online learning of the modified Elman network is realized,and the sensitivity formula to solve the neural generalized predictive control law is presented.Experiments are made on the velocity predictive control of AUVs with on-line and off-line neural network learning,and the predictive control results are analyzed contrastively.Experiment results show that the velocity control accuracy of the AUV with on-line learning is more accurate than that of the AUV with off-line learning,and the former has stronger adaptability when the AUV dynamics changes.
 Yuh J.Design and control of autonomous underwater robots:A survey[J].Autonomous Robots,2000,8(1):7 -24.
 Ishii K,Ura T.Adaptive neural-net controller system for an underwater vehicle[J].Control Engineering Practice,2000,8(2):177-184.
 Soloway D,Haley P J.Neural generalized predictive control:A Newton-Raphson implementation[A].Proceedings of the IEEE International Symposium on Intelligent Control[C].Piscataway,NJ,USA:IEEE,1996.277-282.
 Henriques J,Gil P,Dourado A,et al.Application of a recurrent neural network in on-line modelling of real-time systems[C/OL].http://eden.dei.uc.pt/～ pgil/stuff/esit99 _ 2.pdf.
 Haley P J,Soloway D,Gold B.Real-time adaptive control using neural generalized predictive control[A].Proceedings of the American Control Conference[C].Piscataway,NJ,USA:IEEE,1999.4278-4282.
 Sorensen P H,Norgaard M,Ravn O,et al.Implementation of neural network based non-linear predictive control[J].Neurocomputing,1999,28(1):37 -51.