CVC work roll uneven wear prediction model of hot strip mills based on CNN-BiLSTM-attention + NSGA-Ⅱ

Abstract

The CVC (Continuously Variable Crown) work rolls of hot strip mills suffer from severe uneven wear, and traditional models for predicting work roll wear are difficult to adapt to this uneven wear, resulting in low prediction accuracy. Therefore, a CNN-BiLSTM-Attention + NSGA-II model based on the combination of theory and data is proposed, which can be used for high-precision prediction of CVC work roll wear. The CNN-BiLSTM-Attention intelligent algorithm model combines the advantages of each network layer to achieve higher wear prediction accuracy. The CNN layer comprehensively extracts features from rolling process data, the BiLSTM layer handles the bidirectional time-series characteristics of work roll wear over their service life, and the Attention focuses on key parameters influencing work roll wear. The NSGA-II optimisation model effectively characterises the uneven wear state of CVC work rolls by incorporating wear mechanisms. The wear prediction of CVC work rolls was completed by combining the wear law and the parameters of the rolling mill. The results indicate that the CNN-BiLSTM-Attention + NSGA-II combined model achieves the highest prediction accuracy, with an R² of 0.9736, mean absolute error of 0.0046, and root mean square error of 0.0061. This model provides significant reference value for work roll wear prediction.