【摘要】 钢包精炼炉(LF)钢水硫含量的终点预报对于精确控制精炼成品钢水的成分,稳定钢材质量具有重要意义。针对脱硫过程的机理难建模问题,研究了一种混合模型结构的钢水硫含量终点预报模型,首先基于脱硫反应涉及的物理化学原理建立机理行为已知部分的模型,随后利用正则化网络学习机的自学习能力对机理关系不明确部分进行建模,从而实现了机理方法与数据建模方法的有机结合,取得了较好的预测表现。此外,针对传统正则化网络学习机存在的模型结构复杂度高问题,提出了基于ALD稀疏化算法的模型结构简化方法,该方法可以在不降低预测精度的前提下,显著降低正则化网络学习机的支持向量个数。最后,通过现场实际数据验证了该混合预报模型的有效性。更多还原

【Abstract】 The end-point prediction of sulphur content is very important for accurately controlling the chemical composition of liquid steel in ladle furnace,thereby stabilizing the quality of steel products. Aiming at the difficulty in the mechanism modeling of desulphurization process,In this paper,a hybrid structured end-point sulphur content prediction model is proposed,in which a simplified mechanism model is established based on the physical and chemical principles of desulphurization reaction to capture the known part of the process characteristic,while a data-driven modeling method is introduced using regularization network learning machine to describe the unknown part of the process characteristic. As a result,the advantages of mechanism algorithm and data-driven modeling method can be combined,and better prediction performance is obtained. Moreover,aiming at the high model structure complexity of traditional regularization network learning machine,a model structure simplification method based on ALD sparse representation algorithm is suggested,this method can obviously reduce the number of the support vectors of the regularization network learning machine without deteriorating the prediction accuracy. Finally,the proposed hybrid prediction model is verified by the practical field experiment data.更多还原