An Approach Using Neural-Fuzzy With Modified Fuzzy Associative Memory for System Identification
Hsin-Chi Chang, Wen Feng Lu · Dynamic Systems and Control · 1996
Abstract In this paper, a Neural-Fuzzy with modified fuzzy associative memory (NFFAM) approach is developed for the system identification. This approach contains two mechanisms: modified fuzzy associative memory (FAM) and neural-fuzzy. It combines the learning capability of the neural networks and the linguistic characteristic of fuzzy rules. The modified FAM is used to define the initial structure and initial parameters of the rule base, while neural-fuzzy is used to adjust the rules defined by modified FAM. The simulation results show that for a Mackey-Glass time series prediction problem, the proposed NFFAM itself only implements 61 fuzzy rules to simulate the system, and the learning rate is faster than the Back Propagation Neural Network (BPNN). Besides the advantages of a faster learning rate and self-organization, the NFFAM is a rule-based reasoning approach. Being able to recognize the characteristics of the system from rules was the reason for developing this approach.