A fuzzy inference model (F1M) for plasma shape recognition applications is presented. The model is directly extracted from a data set of examples of the problem without using any learning procedure. The most relevant advantages of the FIM are: 1) the solution of the problem can be expressed in terms of very simple as well as explainable rules, and 2) a very limited number of inputs is required to obtain a sufficient estimation accuracy. The first objective overcomes one of the most limitations of neural network models. The second one has a strong impact on the throughput time in real time applications. The resulting model can be tuned by varying the parameters of the membership functions (centres and variances of the gaussian functions) in order to best fit the data set distribution. The qualitative analysis of the data set may also capture relevant insight on some difficult aspect of the problem, like its basic ill-posedness and the detection of category transition. The results presented in this paper regards a benchmark database of simulated plasma equilibria in the ASDEX-Upgrade machine. The main conclusion is that a FIM is an efficient tool for real time analysis of magnetic data in tokamak reactors.
展开▼
