Type-2Fuzzy Logic System Design And Its Application To Nonlinear System Identification

In recent years, studies on type-2fuzzy logic system (T2FLS) have drawn much attention. The methods of designing a type-2fuzzy logic system and its application to nonlinear system identification are researched in this paper. Firstly, an evolutionary interval type-2TSK fuzzy logic system (EIT2FLS) is developed based on self-evolutionary strategy. Both fuzzy rulebase and parameters are optimized by learning algorithms using training data. And gradient-descent and recursive least-squares with forgetting factor algorithms are implemented to optimize the parameters. Furthermore, a more efficient type-reduction method, enhanced iterative algorithm with stop condition (EIASC), is utilized to reduce the computational cost. The results of applying EIT2FLS to nonlinear system identification issues have demonstrated that the accuracy of the developed EIT2FLS is superior to those of existing methods.Based on the above work, considering the complexity of designing an interval type-2fuzzy logic system (IT2FLS) and the internal relationship between type-1fuzzy system (T1FLS) and IT2FLS, a modified type-2fuzzy logic system learned by a T1FLS, denoted as MT2FLS, is proposed. MT2FLS contains a T1FLS and an IT2FLS which is an extension of the T1FLS. The designing process of MT2FLS mainly includes two parts:fuzzy rulebase learning and parameter learning. For fuzzy rulebase learning, on the basis of self-evolutionary strategy, a new fuzzy rulebase learning algorithm is proposed to learn the rulebase of MT2FLS; first, the rulebase of T1FLS is learned, and then extends the rulebase to get the rulebase of IT2FLS, finally we have the rulebase of MT2FLS. For parameter learning stage, only the parameters of T1FLS in MT2FLS are learned by gradient descent and recursive least-squares with forgetting factor algorithms. The proposed MT2FLS reduces complexity of designing an IT2FLS both in rulebase learning and parameter learning. Compared with existing methods of designing an IT2FLS, the computational cost of MT2FLS will be reduced a lot. The MT2FLS is verified by four nonlinear system identification issues, and the results show that MT2FLS achieves good performance and reduces computational cost.