Research On Neural Generalized Predictive Control Of Autonomous Underwater Vehicle

Autonomous underwater vehicle (AUV) has been widely used in deep sea activities, such as inspection, long range survey and oceanographic mapping, etc. It represents the develop direction of underwater vehicle, and has gotten more and more attention. However, AUV is a highly nonlinear and time-varying dynamic system. Equipped with mechanical arm, AUV becomes a high-level redundant structure, moreover, the uncertain hydrodynamic coefficient also influence the dynamic characteristics of AUV. For these reasons, it is very difficult to control AUV, and a suitable intelligent control technology is exigently desired.To meet these challenges, this paper presents a generalized predictive control system based on modified Elman neural network for an autonomous underwater vehicle. Aiming at identifying high-level and highly nonlinear characteristic of AUV, this paper analyses the system identification algorithm, and a back propagation through time algorithm is applied to train modified Elman neural network. The simulation results of System identification based on multilayer feed forward neural network and modified Elman neural network are compared, the compare results show that the modified Elman neural network is superior. Aiming at the time-varying characteristic of AUV, this paper establishes a neural network on-line model. Considering to the on-line time limitation of AUV computer control system, rolling sample on-line identification algorithm is presented and on-line identification pattern is modified. The structure of neural generalized predictive control system and neural generalized predictive control algorithm are also presented, at the same time, the derivative formula of BP and modified Elman neural network is proposed, the neural network multi-step model is founded. Simulation experiment of generalized predictive control system is studied with restricted control signal and random disturbance based on BP neural network and modified Elman neural network. The simulation results prove that the latter control system is more suitable for AUV. Based on the simulation results, the proposed control system is used to control open-shelf "Beaver" AUV which is developed by Intelligent Control Technology Lab of Harbin engineering university. Several experiments have been carried out, such as, surge and yaw control experiments, the relative close-loop identification experiments and robustness experiment. The experiment results verified the effectiveness of the control system proposed in this paper for AUV with restricted control signal, random disturbance and varying dynamic, moreover, the control system has strong robustness.