The Pertubation Properties And Applications Of A Class Of Fuzzy Hopfield Neural Networks

Artificial intelligence with uncertainty is a hot research area in artificial intelligence domain now. Fuzzy neural networks are important tools to construct nonlinear systems in uncertain knowledge environment.The fuzzy Hopfield networks (Max-T FHNN) which based on Max and T-norms were proposed recently, they have many good properties, not only the robustness of Max-T FHNN to perturbations of training patters but also the practical applications of Max-T FHNN do not start research .So, the two sides have been investigated in the thesis:(1) Theoretic research.When constructing fuzzy neural network systems, the transcendental knowledge like training patterns of systems usually have small perturbation, the perturbation is represented as the errors between training patterns and real patterns,such perturbation of given training patterns probably make various disadvantages to follow processing. Firstly a new concept for the robustness of Max-T FHNN to perturbations of training patterns is proposed. Then combining a new general learning algorithm proposed for Max-T FHNN, the influences of perturbations of training patterns on Max-T FHNN is analyzed. The theoretical studies show that if the Max-T FHNN based on Godel-norm(Godel-norm is a kind of T-norms),then the Max-T FHNN has no good robustness when the general learning algorithm is employed, and if the T-norm and its concomitant implication operators gets the qualification of Lipschitz , then it has good robustness. Then some theoretical results are also confirmed by the experiment on image processing. This work is favorable for the performance analyses, the choice of T-norms and the collection of training patterns for Max-T FHNN.(2) Application research. During the time collecting samples in many pattern recognitions, because of the environmental noise and the precision of equipment impacts, the data that have been collected always have some perturbations, the perturbations have disadvantages to the effect of pattern recognition system. The practical applications of the Max-T FHNN model have been studied for the first time in this thesis. Using Max-T FHNN and other pattern recognition techniques, a vehicle recognition method for intelligent traffic management is proposed. Experiments prove that compared to other method, the recognition rate is better when the method is used under noise environment.