Self-organizing Fuzzy Neural Network Autopilot

The automatic steer is an important equipment in shipping maneuvering. As the increasing need of safety, the requirement for the automatic steer gets greater and greater. From 1920's when American Sperry and German Ansuchz got essential process in the research of Roland and made the mechanical steer, it has been worked over and over. Later on the base of development of electronics and its application, PID steer appeared which integrated control technology and electronic parts. In the late 1960's, adaptive theory and computer technology developed speedily, then adaptive steer came forth by using adaptive theory in shipping maneuvering. Neither PID steer nor adaptive steer shares the character flaws, which results in unassuring effects.The development of the steer was limited until the artificial intelligence (AI) was introduced into control. AI accords with the way of people's thinking and is consisted of fuzzy control, neural network control and their integration. The fuzzy system is designed by imitating the way of fuzzy logical thinking. It can accept blurry variables. On the other way, the reason that the neural network can support fault tolerant in the procedure of computing and dealing information is its special structure. Fuzzy logical reasoning has the ability of expressing structural knowledge, and neural network can self-study. The Fuzzy neural network is the integration of these two advantages.The application of fuzzy neural network controller (FNNC) in automatic steer is mainly discussed in the paper, which integrates the way of fuzzy logical control and the use of neural network. The fuzzy deduction and anti-fuzzy in fuzzy systems are finished by FNNC. Off-line training the FNNC let it study and remember the experiential knowledge. Then the FNNC can simulate the people's control actions. It self-adjusts and self-studies (adjust the degree of FNNC and modifies the control rules) on the base of the object's running status. Finally the control result is improved by the output of FNNC. Another neural network called PNN is introduced to finish constructing and renovating the knowledge database. It can remember performance guides by off-line training. In the procedure of control, PNN studies on-line and gives the signals to FNNC.One of the algorithms to train networks is genetic algorithm which is based on Darwinism.To avoid converge ahead of schedule or enter into the super-plane, the ratio of intercross andaberrance is self-adaptive to enhance the efficiency. On the other hand, Extended Kalman Filter algorithm (EKL) comes into being by the way of improving Kalman Filter.At the end of the paper, programming with VC++ and MATLAB did the simulation of FNNC. FNNC has a lot of advantages by the comparison of PID controller and adaptive controller.But the AI control is only a new thing. It is restricted by hardware in the real application obviously. But as the coming of new theories such as wear structure theory , consultatics, chaos dynamics , ants algorithms etc and the fast development of computer technology, the AI control will have an expansive space in ships.