Quality prediction of continuous annealing process based on HDFI-SVR Model

During the continuous annealing generation process, predicting and monitoring product quality to improve product quality and reduce waste is of great significance. With the improvement of automation, the soft sensor model based on historical process data is also gradually applied to the continuous annealing process. However, the continuous annealing data have the characteristics of high dimensionality, nonlinearity, and strong coupling, which will be a great test for the model establishment. In order to fully consider the impact of data characteristics on modeling, we combine the Isometric Feature Mapping (Isomap) algorithm with the characteristics of integrated learning to further consider the impact of sample differences. At the same time, Support Vactor Regression (SVR) was selected as the sub-learning machine, and then the High-dimensional data features integrated- SVR (HDFI-SVR) model was proposed. Compared with the Adaboost-SVR and SVR models, the model has better convergence speed, generalization speed, and modeling accuracy in different benchmark data sets. After that, the method was applied to build a continuous annealing product quality prediction model, and the experimental results verified the effectiveness of the method. The proposed HDFI-SVR modeling method is more suitable for strip hardness prediction than other test methods.