Parameters optimization of fuzzy-neural dynamic model

In this paper we proposed a fuzzy neural network model which can embody a fuzzy Takagi-Sugeno model and carry out fuzzy inference and support structure of fuzzy rules. The algorithm of model properties improvement consists of new origin procedures namely input space partition, fuzzy terms number and rule number extending, low-effective fuzzy terms and rules extraction and consequent structure identification. In the proposed fuzzy modeling method we first design a rough initial fuzzy model with complete partition of input variable space (or initial partition based on expert knowledge). Then a fuzzy neural network is constructed based on rough fuzzy model. By learning of the neural network we can tune of embedded initial fuzzy model. Next, the additional identifying procedure is introduced based on additional partition of fuzzy input space to improve the properties of initial fuzzy model and to decrease the model error. In final part of identification some low-effective terms and rules are extracted and final rule based model is formed. To apply the new identifying procedures and to introduce possibilities of variability of their properties some parameters have to be put in. The strategy of such parameter optimization is provided by new advanced genetic algorithm. Criterion and cost function has been selected as global fuzzy-neuro model error. To show the applicability of new method and to make a possibility to real systems modeling, we designed the fuzzy-neural network programme tool FUZNET. There were two case studies performed: the first case study presents the prediction of Mackey-Glass time series with using fuzzy-neural regression model (FNRM) predictor; the second case study presents task of a coke-oven gas cooler modeling.