| Full text | |
| Author(s): |
Sussner, Peter
;
Alencar, Tiago da Silva
;
Dick, S
;
Kreinovich, V
;
Lingras, P
Total Authors: 5
|
| Document type: | Journal article |
| Source: | APPLICATIONS OF FUZZY TECHNIQUES, NAFIPS 2022; v. 500, p. 12-pg., 2023-01-01. |
| Abstract | |
Non-Singleton fuzzy logic systems (NSFLSs) have two main advantages over singleton ones: 1. They do not assume that the input data, possibly given by measurements, are precise; 2. They were experimentally shown to have a better capability than conventional singleton fuzzy logic systems to deal with noise. If there is prior knowledge regarding the uncertainty in the input data, e.g., if the measurement error is known, then one is in a position to model non-singleton input fuzzy sets directly, e.g., using intervals. Otherwise, one usually chooses certain types of input type-1 or type-2 fuzzy sets and attempts to estimate their parameters. Usually, one assumes that the training data are noise free. In this paper, we first consider the scenario where both the training and the testing data are corrupted by Gaussian noise with zero mean with an unknown variance. In this case, we smooth the training data using a statistical method before generating the rules. The input fuzzy sets in the testing phase are interval type-2 Gaussians that are centered at the filtered singleton inputs. We propose a method to estimate their parameters. This method also can be applied to the scenario where the testing data are corrupted by arbitrary (stable or varying) levels of noise and the training data is either noisy or noise free. We apply the approaches introduced in this paper to the prediction of chaotic non-linear time series and present some experimental results. (AU) | |
| FAPESP's process: | 20/09838-0 - BI0S - Brazilian Institute of Data Science |
| Grantee: | João Marcos Travassos Romano |
| Support Opportunities: | Research Grants - Research Centers in Engineering Program |