Research On The Neural Network Adaptive Control For Autonomous Underwater Vehicle's Horizontal Tracking Problem

AUV(Autonomous Underwater Vehicle) is a typical system which is nonlinear, coupled and the hydrodynamics of the vehicle are poorly know. When it sails near the water surface, it becomes more difficult for tracking problem because of the immeasurable disturbances. This paper does some deep research on neural network control theory and this theory is applied to AUV heading control system and tracking system.Since linear model of heading control system which is designed based on Taylor series expansion will lead to errors in the system, a nonlinear model of heading control system fitted for the control system design was derived by reasonable simplification of standard submarine 6-DOF motion model, and wave disturbances and parameter uncertainty are added to the model.An adjuster added to neural network when the weights of neural network are modified in the error back-propagation stage decreases network's sensitivity to compensate the slow convergence speed of traditional BP algorithm. This adjuster improves the learning ability of neural network because it changes the omni-directional propagating network into the local back propagating network.Since environment disturbances and the uncertainty of AUV model make a high degree of autonomy difficult to AUV heading control, a dynamic neural network with evaluation factors to the AUV heading control is proposed to avoiding the disadvantage of traditional controller that control performance will become bad when work condition changes. The simulation results shows that the single neural network controller satisfies the maneuvering requirements of AUV in a certain speed to some extent, and gives the better performance of using neural network to control the vehicle in the presence of unpredictable changes in the dynamics of the vehicle and its environment.Since the learning convergence speed of neural network is slow and cannot be much better although it is improved, a hybrid controller is proposed to deal with AUV heading control. This hybrid controller can response the no-model, inaccurate model, disturbances and other uncertainties to the whole control action by just changing a little of the traditional PID control structure. Compared with the single neural network control, that neural network is not a simple replication of the PID controller, although it is trained by the outputs of PID. PID controller is added to enhance system stability, and the controller's main performance is decided by neural network. When disturbances exist, the performance of hybrid controller is better than PID. Especially when the time-variant system has uncertain hydrodynamic coefficients, the hybrid controller can achieve a good control effect through learning stage of neural network. Comparatively, the hybrid controller is of better adaptability, robustness and higher precision.This hybrid controller is applied to AUV tracking control and the tracking curves in the presence of unpredictable changes are obtained. The results show that this hybrid controller improves the dynamic and robustness of AUV control system and has wonderfully potential application.