Research on a feature selection method based on median impact value for modeling in thermal power plants

Highlights: • The concept of median impact value is proposed inspired by mean impact value. • Median impact value method is applied to do feature selection. • The median impact value method can effectively select feature variables. - Abstract: In the paper, the median impact value method similar to the mean impact value method is proposed for feature selection, and the procedure of median impact value method is applied to the prediction model of the main steam flow in thermal power plants, referred to the procedure of mean impact value method. The results show that the sequence of impact values from independent variables to the dependent variables is basically in accord with each other by the median impact value method and the mean impact value method, with only respective independent variables' order changed. The independent variables chosen to set up the main steam flow model are consistent with in the two methods when the accumulative contribution accounts for over 85%. By comparison, the median impact value method is more robust, realistic and effective than the mean impact value method.