基于RST的连退炉温度神经网络模型研究

RESEARCH ON NEURAL NETWORK MODEL FOR CONTINUOUS ANNEALING FURNACE TEMPERATURE BASED ON RST

准确的连续退火炉温度控制是高质量冷轧生产的基本要求,然而变量的不确定性和不完整性以及测量误差的存在使这项任务具有挑战性。人工神经网络在这一领域虽然得到了广泛的应用,但是它们可能无法提供所需的准确性。随着深度置信网络、堆叠自编码机等算法的出现,使得越来越多的学者倾向于对无监督—微调模型的研究。人们寄期望于无监督模型可以挖掘出数据中潜在的关系和知识,然后通过微调引导,从而得到一个更好、更具鲁棒性的模型。基于该思想,将粗糙集理论引入退火炉的神经网络“遗忘门”部分,通过粗糙集理论决定变量的淘汰和保留。

Accurate continuous annealing furnace temperature control is a fundamental requirement for high-quality cold rolling production,but the uncertainty and incompleteness of variables,as well as the presence of measurement errors,make this task challenging.Although artificial neural networks have been widely used in this field,they may not provide the required accuracy.With the emergence of algorithms such as deep confidence networks and stacked self coding machines,more and more scholars are inclined to study unsupervised fine-tuning models.People expect unsupervised models to uncover potential relationships and knowledge in the data,and then fine-tune guidance to obtain a better and more robust model.Based on this idea,this article introduces rough set theory into the"forgetting gate"part of the neural network of the annealing furnace,and determines the elimination and retention of variables through rough set theory.