Singular value-based fuzzy reduction with relaxed normality condition

The paper addresses an issue of reduction of two-input-single-output fuzzy rules whose condition parts are composed of numerical singletons \(r_{ij}\). Given \(n_1\) and \(n_2\) fuzzy sets defined in the corresponding input spaces, these singletons are arranged into a matrix form \(T= [r_{ij}]\), \(i= 1,2,\dots, n_1\), \(j= 1,2,\dots, n_2\) to which one applies the singular value decomposition meaning that \(T\) transforms into the expression \(T =URV^T\) with \(\dim(U)= n_1\times n_1'\), \(\dim(R)= n_1'\times n_2'\), \(\dim(V)= n_2\times n_2'\). Based on the SVD, the authors form a new rulebase along with modified fuzzy sets of antecedents. The number of reduced rules depends upon the number of singular values retained in the SVD process. As the original fuzzy sets occurring in the rules become also affected by the SVD, the authors propose a set of conditions for membership functions that retain their interpretability. Those conditions include: (a) sum normalization, (b) nonnegativeness, and (c) normality. Illustrative numerical examples are presented as well. For the entire collection see [Zbl 1015.00012].

Recommendations

Cited in

(7)

This page was built for publication: Singular value-based fuzzy reduction with relaxed normality condition

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4437867)