Fuzzy neural network based on rectangle functions and its application

Fuzzy neural network (FNN) based on rectangle functions is constructed by partitioning input space into many disjoint hyper-cubes with the same size. FNN is constant in each of the hyper-cubes. If and only if an input sample drops into a hyper-cube would the corresponding sample be memorized through coding. Moreover, FNN can generate fuzzy rules automatically. For the control of a nonlinear system, a theorem about static error shows that static error can be decreased for small enough partition of the input space. Simulation example shows that result is satisfactory.