A FNN Control of Underwater Vehicles based on Ant Colony Algorithm

For the particular controlled object AUV, a novel controller based on the fuzzy B-Spline neural network is presented, which embodies the merits of qualitative knowledge representation capability of fuzzy logic, quantitative learning ability of neural networks, as well as the excellent local controlling ability of B-Spline basis functions. However, to overcome the inherent deficiencies in the fuzzy neural network, including the structure hardly to be fixed, slow-speed training with the tendency to be involved in local convergence, and the quality of training results dependent upon the initial conditions of the network as well, some optimizing efforts are carried out in this investigation. The improved dual ant algorithm is employed for offline optimization, which can efficiently avoid the phenomenon of precocity and stagnation during the evolution. Meanwhile, the expert experience is introduced to simplify the number of optimizing parameters, and then the controller is further improved with the hybrid training by adopting the BP algorithm proceeding online adjustment. The simulation of the AUV motion control demonstrates the feasibility and validity of the present method.