1. Deep stochastic configuration networks with different distributions for crown prediction of hot-rolled non-oriented silicon steel
- Author
-
Ding, JingGuo, Du, HaoZhan, Meng, LingMing, Zhao, Jian, Wang, GuiQiao, and Zhang, DianHua
- Abstract
In the hot rolling process, strip shape is a great factor in assessing the quality of non-oriented silicon steel products. As the downstream industry has increasingly quality requirements for silicon steel products, the traditional predictive model is unable to meet the demand for silicon steel control. In order to improve the accuracy of the pre-set crown model, a stochastic configuration network (SCN) is used to build a data-driven crown prediction model. Additionally, we conduct correlation analysis of all features using Spearman correlation analysis to mitigate the influence of redundant variables. To improve the network's performance, we investigated the effect of weights and biases on SCN performance in four different distributions. We utilized a zero-symmetric distribution with adaptive parameters to decrease the model's training time and enhance its generalization performance. The results indicate that the model achieved the highest prediction accuracy under the normal distribution of adaptive parameters, with the correlation coefficient of 0.9483. The accuracy of crown deviation within ±5 μm is 96.3 %, the root mean square error and the mean absolute error are 2.525 and 1.925 respectively. Furthermore, the model prediction results were used to analyze the impact of four key rolling parameters on the crown of silicon steel. Compared to the original SCN, this method can effectively enhance the prediction accuracy of the silicon steel crown.
- Published
- 2024
- Full Text
- View/download PDF