Dynamic underwater robot recurrent neural network control with Petri-threshold

Abstract

Traditional fuzzy network controller was disadvantadged in heavy caculation and hysteresis response to strong disturbance.Therefore a fuzzy recurrent neural network controller was designed,in order to improve the robustness corresponding to environment change through online dynamic feedback.Threshold was issued in the third layer so as to regulate training and learning according to controller errors.Thus caculation of the whole network was reduced.Moreover,the online training algorithm was developed based on gradient descent method.The learning rate parameters were determined according to discrete-type Lyapunov function,which guaranteed the whole network convergence.Experiments have demonstrated that the controller can improve the computation efficiency,reduce control errors,possess strong robustness and be very effective in the underwater robotic control.