基于鲸鱼算法优化BP神经网络的结晶器液面波动的预测  

Prediction of mold level fluctuation based on BP neural network optimized by whale optimization algorithm

作　　者：徐猛 刘娟娟 雷洪 李强 张秀香 XU Meng;LIU Juanjuan;LEI Hong;LI Qiang;ZHANG Xiuxiang(Key Laboratory of Electromagnetic Processing of Materials,Ministry of Education,Northeastern University,Shenyang 110819,China;School of Metallurgy,Northeastern University,Shenyang 110819,China;Benggang Steel Plates Company Limited,Benxi 117000,China)

机构地区：[1]东北大学材料电磁过程研究教育部重点实验室,辽宁沈阳110819 [2]东北大学冶金学院,辽宁沈阳110819 [3]本钢板材股份有限公司,辽宁本溪117000

出　　处：《冶金自动化》2023年第2期66-72,共7页Metallurgical Industry Automation

基　　金：国家自然科学基金联合基金项目(U1460108)。

摘　　要：在连铸过程中,结晶器液面波动是限制连铸速度和铸坯质量的关键参数之一,因此,液面波动行为的准确预测一直是冶金学者的研究重点。基于此,本文利用Python对结晶器液面波动的振幅值进行预测。首先,选取中间包的质量、塞棒的位置、拉力和结晶器振动作为模型的输入,对数据快速傅里叶变换和归一化处理。然后,构建4×3×1的反向传播(back propagation,BP)神经网络模型,并利用鲸鱼算法(whale optimization algo-rithm,WOA)对初始权值和阈值优化。通过训练预测,相比BP神经网络,WOA-BP神经网络能较好地对结晶器液面波动进行预测,且预测值与结晶器液面波动振幅吻合较好,拟合决定系数(R^(2))为0.8414;当振幅偏差为±0.02时,命中率可达到91%。In the continuous casting process,the fluctuation of mold level is one of the key parameters to limit the casting speed and the slab quality.Therefore,the accurate prediction of liquid level fluctuation behavior has always been the focus of metallurgical scholars.Based on this,Python was used to predict the amplitude value of the mold level fluctuation in the paper.Firstly,the weight of the tundish,the position of the plug rod,the withdraw force and the vibration of the mold were selected as the input of the model,and the data were processed by fast Fourier transform and normalization.Then,a 4×3×1 back propagation(BP)neural network model was constructed,and the initial weights and thresholds were optimized by whale optimization algorithm(WOA).Through training prediction,the WOA-BP neural network can predict the fluctuation of mold level better than the BP neural network,and the predicted value matches the amplitude of mold level fluctuation,and the Rsquared(R^(2))is 0.8414.When the amplitude deviation is±0.02,the hit rate reaches 91%.

关 键 词：BP神经网络 连铸 结晶器 液面波动 鲸鱼算法 

分 类 号：TF777[冶金工程—钢铁冶金] TP18[自动化与计算机技术—控制理论与控制工程]