Fan Rotation Speed Real-Time Optimizations of Continuous Annealing Line with Mechanism-Guided Multitask Classification and Regression Model

Abstract

Abstract Fast cooling with gas jets in the rapid cooling section plays significant role in the process of steel strip galvanization. The large number of the cooling fans to generate the gas jets and the frequent change of the strip dimension, the strip velocity and the inlet strip temperature contribute to the great complexity of the fan rotation speed regulations. The experience-dominated regulation method does not work well in real production. This paper proposes a multitask classification and regression (MTCR) model to optimize the fan rotation speed in real time. A heat transfer model is firstly built in the form of partial differential equations (PDEs) and is used to construct the features of the MTCR model. The overall heat transfer coefficient is calculated and analysed. More than 575, 000 data records from the real production line are used to construct the features, and train and evaluate the MTCR model. In addition, the predictions of the MTCR model are compared with that of the multitask regression (MTR) model and the single task regression (STR) model. All the models perform well for the rotation speed predictions of the switched-on fans, while the MTCR model performs better in the whole test dataset.