Research On The Optimal Design Of Data-driven Type II Fuzzy Systems

The type-2 fuzzy system has been widely used in the fields of system modeling,control and time series forecasting due to its advantages in dealing with highly uncertain problems.In order to construct a good-performance type-2 fuzzy model,data-driven strategies are usually used to optimize the design.However,the optimization of type-2 fuzzy model faces a series of problems such as high computational complexity,long parameter optimization time,and relatively slow model learning speed.In this thesis,aiming at the optimization design problem of the data-driven interval type-2 fuzzy system,a data-based rule reduction method of the interval type-2 fuzzy model is given and the model parameter learning strategy based on the differential evolution algorithm is designed;the optimization construction method of the interval type-2 fuzzy model based on the distributed integration strategy is proposed.The specific content and contributions of this thesis are as follows:First,the basic knowledge of fuzzy model is introduced.The definitions of type-1 and type-2 fuzzy sets,fuzzy set operations and fuzzy model reasoning structure are described respectively.The focus is on the fuzzy reasoning process of the interval type-2 fuzzy model,and the advantages of type-2 fuzzy in dealing with highly uncertain problems are analyzed.Then,a design method of interval type-2 fuzzy model(T2FM-DE)based on differential evolution and rule reduction is given.This method firstly adopts a data-driven method to reduce the complete rule base of the model.The reduction reduces the number of fuzzy rules to a certain extent and reduces the scale of model parameters.On the basis of rule reduction,differential evolution algorithm is used to optimize the parameters of the reduced interval type-2 fuzzy model to further improve the model performance.In order to verify this method,experiments are carried out on two prediction problems of wind power generation and subway pedestrian flow.And compared with mainstream methods,including BP neural network(BPNN),adaptive fuzzy inference system(ANFIS),etc.,the experiment and comparison results show that the proposed T2FM-DE method has good prediction performance in prediction problems.Secondly,in order to improve the learning speed of the interval type-2 fuzzy model,a data-driven method of distributed integration construction of the interval type-2 fuzzy model is proposed.In this method,the original training data set is divided into multiple data subsets,and a corresponding type-1 fuzzy model is constructed for each data subset by using a distributed parallel processing mechanism;the model integration idea is adopted to integrate the constructed type-1 fuzzy model to obtain the initial interval type-two fuzzy model;the overall parameters of the interval fuzzy-2 model's subsequent parts are optimized based on the least squares method to further improve the model's prediction performance.Experiments and comparative analysis are carried out on wind power generation and subway passenger flow forecasting problems through models such as T2FM-DE,BPNN and ANFIS.The results show that the proposed method of distributed ensemble construction of interval type-2 fuzzy models has a faster learning speed on the basis of ensuring the performance of the model.Finally,this thesis analyzes and summarizes the designed interval-type two fuzzy model,and reflects on the deficiencies in the methods and models,and on this basis,prospects for future research directions.