摘要
In view of the limitations in the prediction of pollution flashover voltage by least squares regression, a method to predict pollution flashover voltage by robust regression is proposed. According to testing voltage and the data of salt deposit density (ρSDD ) and non-soluble deposit density (ρNSDD ), the regression coefficient is solved by a complex weighting least square iteration algorithm. In iterative calculations, the weight function is adopted, in which the weight coefficient is the function of the residual error of last iteration to weaken the influence of singular values on the regression coefficient. The characteristic exponent denoting ρSDD influence and characteristic exponent denoting ρNSDD influence are mapped by the regression coefficient, and thus the pollution flashover voltage of insulators can be predicted. Through the comparison of test results, robust regression results and least squares regression results, the effectiveness of the proposed robust regression-based forecasting method is verified.
In view of the limitations in the prediction of pollution flashover voltage by least squares regression, a method to predict pollution flashover voltage by robust regression is proposed. According to testing voltage and the data of salt deposit density (ρSDD ) and non-soluble deposit density (ρNSDD ), the regression coefficient is solved by a complex weighting least square iteration algorithm. In iterative calculations, the weight function is adopted, in which the weight coefficient is the function of the residual error of last iteration to weaken the influence of singular values on the regression coefficient. The characteristic exponent denoting ρSDD influence and characteristic exponent denoting ρNSDD influence are mapped by the regression coefficient, and thus the pollution flashover voltage of insulators can be predicted. Through the comparison of test results, robust regression results and least squares regression results, the effectiveness of the proposed robust regression-based forecasting method is verified.
基金
supported by Key Scientific and Technical Funds of Zhejiang Electric Power Corporation under Grant ZDK069-2010
