Data-driven modeling by gaussian membership based sample selection and its application in steel energy system

Due to the data diversity and complexity in industrial system, the accuracy of data-based modeling might be largely affected by such a series of issues. Aiming at the energy system in steel industry, this study proposes a fuzzy modeling based on Gaussian membership expression. First, in the stage of sample selection, the industrial data set is divided into a number of clusters, from which the representative sample are chosen based on a variable step rule. Second, given the industrial data usually accompany with high level noise and anomaly points, a fuzzy modeling based on Gaussian membership is proposed, where a sample reliability coefficient is introduced to alleviate the negative impact produced by ill-posed data, and the model parameters solution is explicitly derived later. The proposed method has been applied to the practice of gas flow prediction in a steel plant. To verify its performance, a number of experiments are conducted by using the data coming from the energy center in the plant. The results indicate that the proposed method greatly improves the prediction accuracy and efficiency, which plays a significant role in data-based modeling for the industrial system.