In this paper, we first design a fuzzy neuron which possesses somegenerality. This fuzzy neuron is founded is founded by replacing theoperators of the traditional neuron with a pair of abstract fuzzyoperators as (＋,●) which we call fuzzy neuron operators. Forexample, it may be (＋,●), (∧,●),(∨,●), or (∧,∧), etc. It isan extended fuzzy neuron, and a network composed of such neurons isan extended fuzzy neural network. Then we discuss the relationshipbetween the fuzzy neuron operators and t-norm and t-conform, andpoint out fuzzy neuron operators are based on t-norm but much widerthan t-norm. In this paper we will emphatically discuss a two-layerednetwork and its training algorithm which will have to satisfy a setof various operators. This work is very related to solving fuzzyrelation equations. So it can be used to resolve fuzzy relationequations. Furthermore, the new fuzzy neural algorithm is found to bestronger than other existing methods to some degree. Some simulationresults will be reported in detail.
展开▼
