Random plant identification by a nonparametric approach

A technique is described for obtaining regression estimates with a priori uncertainty on the structure of the stochastic plant model. Nonparametric Parsen estimates are used for simultaneous probability densitities with optimal kernels and fuzziness coefficients. The optimization is performed with the least mean square integral (for inputs) deviation of the regression estimate from its theoretical value as the criterion. The final adjustment of fuzziness coefficients is to meet the requirement of the best smoothing of initial plant input and output measurements. Examples are given.