论文部分内容阅读
钢铁企业连续退火生产过程由于工艺复杂并且包含众多相互影响的过程变量,导致带钢产品质量波动较大,难以进行在线测量。针对该问题,提出了一种基于混合集成学习的带钢产品质量在线预报方法。该方法以Bagging为基础,引入了Adaboost中对误差较大样本的重点学习策略,以提高混合集成学习模型的精度并改进泛化能力;在子学习机的训练中提出了基于动态加权的最小二乘支持向量机方法,以改进子学习机的鲁棒性。基于实际连退生产过程数据的测试结果表明,所提出的混合集成学习方法在预测精度和泛化能力上均要优于Bagging和Adaboost等传统的集成学习建模方法。
The continuous annealing process in the iron and steel industry due to the complicated process and the numerous process variables that affect each other leads to large fluctuations in the quality of the strip steel and makes it difficult to conduct online measurements. Aimed at this problem, an online forecasting method of strip quality based on hybrid integrated learning is proposed. Based on Bagging, this method introduces the key learning strategy of larger samples in Adaboost to improve the accuracy of the hybrid integrated learning model and improve the generalization ability. In the training of sub-learning machine, Multiply support vector machine method to improve the robustness of sub-learning machine. The test results based on the actual production process data show that the proposed hybrid integrated learning method is better than the traditional integrated learning modeling methods such as Bagging and Adaboost in the prediction accuracy and generalization ability.