Parameter identification of adaptive fuzzy-nerual model with binary-tree structure using genetic algorithm

摘要

An adaptive fuzzy-neural model of a process is presented in this paper. A single-input single-output process is considered, anda collection of measurement data is provided. It is assumed that the process kth sampled output is a function of d consecutive previous inputs and outputs. With d being power of 2, the output forms a binary tree structure function of fuzzy rule based. By neuralizing the fuzzy system, this structure can be modified into a binary-tree structure of atificial neural network with two hidden layers. Two case studies are demonstrated using box and jenkins gas furnace data. The first model, with linear activation function, required 7 parameters to the identified. The learning process is conducted by using genetic algorithm. In the second model. the adaptive fuzzy neural, we use two activation functions in the hidden layers, from which 9 parameters have to be identified.