Adaptive control based on neural fuzzy inference network
Ion Dumitrache, Nicolae Constantin · 2001
The long training time of multilayered backpropagation neural networks (BPNN) represents a serious drawback for the applications in industry. Moreover when they are trained on-line to adapt to plant variations, the overtuned phenomenon occurs. In this paper a novel neural fuzzy network (NFN) it is proposed which is suitable for adaptive control. The NFN represent a modified Takagi-Sugeno-Kang (TSK) type fuzzy rule based model with neural network learning ability. The rules are created and adapted in an online learning algorithm. The structure learning together with the parameter learning forms the learning algorithms for the neural fuzzy network. It is proved that NFN can greatly reduce the training time, avoid the over-tuned phenomenon and has perfect regulation ability.