Prediction of Unburned Carbon in Flue Dust based on Teacher-student Ensemble Method. Issue 1 (February 2021)

• Record Type: Journal Article
• Authors: Lu, Kuan; Yang, Xingsen; Wang, Haichao; Liu, Ke; Zhang, Xuhui; Meng, Xiangrong; Li, Jun
• Is Part Of: IOP conference series. Volume 634:Issue 1(2021)
• Publication Date: 2021-02
• Subjects: Earth sciences -- Periodicals; Environmental sciences -- Congresses; Environmental sciences -- Periodicals; 550.5
• DOI: 10.1088/1755-1315/634/1/012030
• Languages: English
• ISSNs: 1755-1307

Abstract

Abstract: Based on the idea of knowledge distillation, a teacher-student ensemble method is proposed for UCFD (Unburned Carbon in Flue Dust) prediction when influencing factors have different sampling frequency. First, a multi-layer feed forward network is built as the teacher model. Loss function is customized to account for the sampling period discrepancy between outputs and inputs. With this teacher model, UCFD could have the same rate with its influencing factors. Second, Xgboost and Adaboost are used to form an ensemble student model to improve the training process and prediction robustness. Third, a power plant in Shandong province is chosen to make data experiment. Results illustrate that the teacher-student ensemble method can give a more accurate forecast.