带钢热镀锌镀层单位面积质量神经网络预测模型

Prediction models of coating mass per unit area for hot-dip galvanized strip based on artificial neural network

为了解决带钢连续热镀锌镀层单位面积质量偏差大、调节时间长、锌原料浪费等问题,利用生产过程历史数据建立了气刀压力预设定模型和镀层单位面积质量预测模型。根据带钢热镀锌生产实践,分析了镀层单位面积质量的影响因素和控制策略。收集系统稳态样本数据并进行多变量偏相关分析,证明镀层单位面积质量与气刀压力、气刀距离、带钢速度相关。根据样本数据,以镀层单位面积质量、气刀距离、带钢速度为输入变量,采用BP神经网络建立气刀压力预设定模型,预设定精度达到3000Pa。收集热镀锌过程时间序列样本数据,采用NARX动态神经网络建立镀层单位面积质量预测模型,预测精度达到6g/m^2,为实现带钢热镀锌镀层单位面积质量闭环控制奠定了基础。

In order to solve the problems of large mass per unit area deviation,long adjustment time,and waste of zinc raw materials for continuous hot-dip galvanizing of steel strip,an air knife pressure preset model and a coating mass per unit area prediction model were established using historical data of the production process.Based on the production practice of the hot-dip galvanizing of steel strip,the influence factors and the control strategy of the coating mass per unit area were analyzed.The static sample data was collected,and the multi-variable partial correlation analysis was carried out,which showed that the coating mass per unit area was related to the air knife pressure,the air knife distance and the steel strip speed.According to the sample data,taking the coating mass per unit area,the distance of air knife and the speed of strip steel as input variables,BP neural network was used to establish the preset model of air knife pressure,and the preset precision reached 3000 Pa.The sample data of time series during the hot-dip galvanizing process was collected and a prediction model of coating mass per unit area was established using NARX dynamic neural network.The prediction accuracy reached 6 g/m^2.The foundation for closed-loop control of the coating mass per unit area for hot-dip galvanized strip was provided.