Data And Knowledge Driven Design Of Fuzzy Inference System And Applications

Since the appearance of the definition of fuzzy set,it has attracted more and more attentions in different research areas due to its powerful ability to utilize human or experts' knowledge and its capability to tackle different kinds of uncertainties.Especially,it has found lots of applications in the modeling and control areas,where the fuzzy inference systems are widely used to model the unknown systems or utilized as controllers.However,the fuzzy inference systems inevitably face the fuzzy rule explosion problem when the system to be modeled or controlled has a large amount of variables.In this situation,it will be quite difficult to set up the fuzzy rules and to optimize the parameters of the fuzzy inference systems.To solve the above problems,this paper presents a data and knowledge driven design approach for fuzzy systems.This thesis mainly contains the following contents.Firstly,the structure and monotonicity of the single-input rule module connected fuzzy inference system(SIRM-FIS)are introduced in detail.On this basis,a data and knowledge driven design approach for the SIRM-FIS is proposed.This method uses a data-driven parameter learning strategy to optimize its parameters under constraints that are formed from the knowledge.The system has been applied to the thermal comfort prediction.Simulation and comparisons illustrated that the proposed method is efficient to the thermal comfort prediction and performs better than some other existing methods.Secondly,the functional-type single input rule module connected fuzzy inference system(FSIRM-FIS)which is an extended structure of the SIRM-FIS is introduced.Then,a neural structure is added to the FSIRM-FIS to form a functional-type single input rule module connected neural fuzzy system(FSIRMNFS)to combine the merits of both the FSIRM-FIS and the neural network.At the same time,in order to achieve both the smallest training error and the smallest parameters,a least squares method based parameter learning algorithm isproposed for the FSIRMNFS.The proposed FSIRMNFS and its parameter learning algorithm are applied to the hourly wind speed prediction.Simulation and comparison results have verified the effectiveness of the hourly wind speed prediction.Finally,an ensembling approach based data-driven method is proposed to construct type-2 fuzzy logic systems through merging type-1 fuzzy logic systems generated by the adaptive neural fuzzy inference system(ANFIS)method.The type-2 fuzzy logic system is applied to the wind speed prediction problem.Simulation and comparison results show that,compared with the well-known BPNN and ANFIS,the proposed method have similar performance but greatly reduced training time.With the exponential growth of data,it can effectively reduce the modeling time of type-2 fuzzy logic system.