平整机轧制力的神经网络预报模型研究

发布时间：2018-01-21 23:40

本文关键词： 平整机轧制力神经网络 ReLU 传播算法正则化　出处：《河北工程大学》2017年硕士论文　论文类型：学位论文

【摘要】：钢铁工业支持了国民经济以及国防建设的发展,同时各行各业的发展又推动着钢铁行业的产品质量不断进步。对于提高带钢产品的质量,平整是其不可或缺的一个环节。它不仅能提高带钢的表面质量,还直接影响其物理、化学和力学性能,进而达到后续工艺阶段要求的规格。针对平整机轧制力的预报研究是合理优化平整轧制过程、提高平整机控制水平和改善工作状态所面临的一个重要课题。本文针对平整机轧制力预测精度不高的问题,以影响平整机轧制过程的参数为研究对象,以轧制理论和神经网络为理论依据,提出以ReLU为激活函数的人工神经网络模型来对平整机的轧制力进行预报研究。进行了以下研究工作:对在线轧制数据进行主成分分析降维处理,获得影响平整机轧制力的主要因素,并将其作为主成分输入神经网络模型的输入层神经元,将平整机的轧制力作为神经网络的输出层神经元,以网格搜索的方式对神经网络隐层的相关参数和算法进行实验,采用python语言进行编程,建立了2360组平整机轧制力的神经网络预报模型。基于上述研究内容和成果,利用建模分析,结合大量现场平整轧制数据的分析处理,通过调整隐层层数、神经元数、传播算法、正则化方法,筛选出了预测误差最低的神经网络模型。同时,这种实验方法可以适用于不同在线轧制数据下的平整机轧制力的预报,对于平整生产具有一定的指导意义与参考价值,同时该实验思路可以推广到其它参数的预报研究中。
[Abstract]:The iron and steel industry has supported the development of national economy and national defense construction. At the same time, the development of various industries has promoted the continuous progress of the steel industry product quality. Flatness is an indispensable part of the strip. It can not only improve the surface quality of strip, but also directly affect its physical, chemical and mechanical properties. According to the prediction of rolling force, it is reasonable to optimize the rolling process. It is an important task to improve the control level and improve the working state of the mill. In this paper, the parameters that affect the rolling process of the mill are taken as the research object, aiming at the problem that the prediction accuracy of the rolling force is not high. It is based on rolling theory and neural network. An artificial neural network model with ReLU as the activation function is proposed to predict the rolling force of the temper mill. The following research work is carried out: the principal component analysis (PCA) dimensionality reduction processing of the rolling data on line is carried out. The main factors that affect the rolling force are obtained and used as the input layer neuron of the neural network model and the rolling force of the mill as the output layer neuron of the neural network. The related parameters and algorithms of the hidden layer of neural network are experimented with the method of grid search, and python language is used to program. The neural network prediction model of rolling force of 2360 sets of leveling mill is established. Based on the above research contents and results, the model is used to analyze and deal with a large number of rolling data, and the hidden layer number is adjusted. The neural network model with the lowest prediction error is selected by neuron number, propagation algorithm and regularization method. At the same time, this experimental method can be used to predict the rolling force of flat mill under different on-line rolling data. It has certain guiding significance and reference value for leveling production, and the experimental idea can be extended to the prediction of other parameters at the same time.
【学位授予单位】：河北工程大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG333.4

【参考文献】