Research On Knowledge Utilization-based Fuzzy Systems And Applications

In recent years,with the progress of science and technology,machine learning technology has developed rapidly.We can see the shadow of machine learning in many fields,such as industrial manufacturing,medical diagnosis,driverless and so on.With the development of economy and technology,the application scenarios of machine learning are expanding.At the same time,it also makes the data form more complex: for example,the data contains more noise or some wrong content,which brings great challenges to the application of traditional machine learning technology.Among many machine learning technologies,Takagi Sugeno Kang(TSK)fuzzy system,as a common classification model,is favored by many scholars because of its excellent nonlinear approximation performance and interpretability.However,when TSK is faced with complex data scenarios,the classification performance and generalization of the model are not ideal.The integrated TSK fuzzy classification model usually has better classification performance than the classical TSK,which can alleviate this challenge to some extent,but at the expense of the interpretability of the model.In order to solve these problems,based on the principle of knowledge utilization,this paper studies the fuzzy system.The main work is as follows:(1)A TSK transfer learning fuzzy system(TSK-TL)is proposed.Based on the interpretable TSK fuzzy system,tsk-tl balances the distribution distance between the source domain and the target domain data when learning the model through the joint distribution adaptive strategy of transfer learning,so that the model can make more effective and reasonable use of the information of the source domain and the target domain.At the same time,it trains new subsequent parameters through the historical knowledge learning mechanism,so as to make the obtained model more effective.The experimental results show that tsk-tl can effectively identify the target samples in the data drift scenario,improve the classification accuracy and have stronger generalization ability.(2)Based on the stacked generalization principle of ensemble learning,this paper further proposes a deep TSK fuzzy system with stacked structure and rule optimization mechanism(STSK-RO).STSK-RO integrates several classical first-order TSK fuzzy systems by stacking,each of which has its own different input and output;At the same time,different TSK fuzzy systems are connected,that is,the output of the previous layer participates in the input of the next layer,thus forming a depth structure.The model also contains three mechanisms related to depth structure: rule transmission,rule forgetting and rule mutation.Under the action of these three mechanisms,the model can continuously learn and optimize the rule structure in different levels,so as to obtain more general knowledge.At the same time,we prove through mathematical transformation that some outputs of the middle part of the model are hidden in the consequent,thus ensuring that the model still has good interpretability.The experimental results show that stsk-ro has strong anti noise ability,can fully mine the knowledge hidden in the data,and eliminate the interference information and error information in the data,so as to obtain good classification performance.