Fuzzy gaussian potential neural networks using a functional reasoning

This paper presents the principal design of a fuzzy gaussian potential neural network (FGPNN) to achieve high capability to learn expert control rules of the fuzzy controller. In this construction, each membership function consists of a gaussian potential function (GPF) which causes the utilization of a reduced number of labels, and eventually the complexity of structural design becomes simple, specially for large scale inputs, This in turn reduces the learning trials, to improve the learning speed. Thus, the time of the training process, which is based on the ba~k-propagation method, is shortened. The construction of an FGPNN is carried out with the minimum number of GPF, based on the number of input patterns, to learn the mean vectors and shapes of the individual GPFs that basically depend on the desired trajectory. Finally, we provide a simulation to evaluate the proposed method for a multi input-output, twolink manipulator.