Control of Instantaneous Abnormal Mold Level Fluctuation in Slab Continuous Casting Mold Based on Bidirectional Long Short-Term Memory Model

The instantaneous abnormal mold level fluctuation (IAMLF) has significant harmful effects on slab quality. This article proposes a prediction and control method for IAMLF. First, the data processed by difference method is used for the IAMLF prediction and stopper-rod position prediction. Then, the bidirectional long short-term memory (BI-LSTM) is introduced to predict the IAMLF; the corresponding stopper-rod position is predicted according to the mold level prediction result. BI-LSTM can predict the IAMLF with the mean absolute error of 1.52 mm and the false alarm rate of 1.8%, and also performs well in predicting the stopper-rod position with a mean absolute error of only 1.92 mm. Furthermore, the two prediction models are combined to form a closed loop, where the mold level fluctuation is predicted according to the industrial data processed by data difference method, and the stopper-rod position is adjusted in advance to eliminate IAMLF. Finally, the fuzzy proportional-integral-derivative (PID) controller is used to control mold level based on the mold level and stopper-rod position prediction results. The prediction accuracy of IAMLF reaches 98.4%, and the present proposed fuzzy PID controller can effectively prevent the occurrence of IAMLF with a success rate of 95.6%.