Compensatory Neural Fuzzy Network And Its Applications In EDM

Nowadays, neural networks and fuzzy logic systems have been becoming the focus of the researches in the field of intelligent control. Once the two techniques are combined together, the Neural Fuzzy Network (NFN) was born, which has the advantages of both of them. Not only can it express approximate and qualitative knowledge, like Fuzzy Logic, but also it has the strong ability of learning and expressing non-linearity. More importantly, its structures of networks have clear physical interpretation to users.However, there exist two problems in the process of designing the traditional NFN as follows:(1) It is not easy to determine the initial fuzzy model, including the number of fuzzy rules, the initial parameter values in the premise part and the consequent part. For the average NFN, the structure of its fuzzy model is usually determined by expert knowledge. However, sometimes it is difficult to extract the full rules, or it can't obtain the expected results because of the variable dynamic characteristics of the objects and the influence of disturbance.(2) The NFN systems used widely now are limited to deal with the static problems, because of the absence of recurrent structure in them. In fact, most processes are always dynamic and the variables of them are the functions of time. Therefore, it is very practical to study the NFN with dynamic characteristics.In order to overcome the first problem, a modified relational grade clustering (MRGC) method is presented, and then design a Compensatory Neural Fuzzy Network (CNFN) based on the MRGC method and Gradient Descent (GD) method.