Kernel based nonlinear fuzzy regression model

Recent years have seen a surge of interest in extending statistical regression to fuzzy data. Most of the recent fuzzy regression models have undesirable performance when functional relationships are nonlinear. In this study, we propose a novel version of fuzzy regression model, called kernel based nonlinear fuzzy regression model, which deals with crisp inputs and fuzzy output, by introducing the strategy of kernel into fuzzy regression. The kernel based nonlinear fuzzy regression model is identified using fuzzy Expectation Maximization (EM) algorithm based maximum likelihood estimation strategy. Some experiments are designed to show its performance. The experimental results suggest that the proposed model is capable of dealing with the nonlinearity and has high prediction accuracy. Finally, the proposed model is used to monitor unmeasured parameter level of coal powder filling in ball mill in power plant. Driven by running data and expertise, a strategy is first proposed to construct fuzzy outputs, reflecting the possible values taken by the unmeasured parameter. With the engineering application, we then demonstrate the powerful performance of our model.

► The so-called kernel based nonlinear fuzzy regression model is proposed by introducing the strategy of kernel into fuzzy regression.
► The kernel based nonlinear fuzzy regression model is a general model that can deal with nonlinearity existing between crisp inputs and fuzzy output.
► The simulation studies show that the proposed model has high predication accuracy.
► The engineering application to monitoring unmeasured parameter in power plant suggests that the prediction fits well with the expertise analysis.