Segmented Mechanism-Data Hybrid Modeling for Outlet SO₂ Mass Concentration Prediction in WFGD Systems Under Flexible Peak-shaving Conditions

Considering the dynamic change of key operating parameters during flexible peak-shaving operation process of coal-fired units, a modeling approach for outlet SO₂ mass concentration in WFGD systems was proposed, which integrated a segmentation strategy with mechanism-data hybrid modeling. Based on the analysis of variation characteristics of key operating parameters during peak-shaving operation process, the dataset was divided into deep and basic peak-shaving intervals, within which mechanism models, temporal convolutional network-long short term memory (TCN-LSTM) models, and hybrid models were built. Results show that segmented modeling strategy has significant advantages in enhancing modeling accuracy of outlet SO₂ mass concentration and has good adaptability to different modeling approaches. The mechanism-data hybrid model achieves the best performance, with R_MAPE and R_MAE reduced by up to 33.4% and 42.1%, respectively, compared with original full-condition modeling.