摘要
针对钢包连铸过程中需要精确控制下渣时刻的问题,提出一种基于局部加权回归和长短时记忆(LSTM)神经网络模型的连铸下渣预测系统。该系统对下渣过程中采集到的信号进行处理和识别,可准确预测下渣时刻。结合某钢厂的实际生产情况,在采集到的大量钢包下渣相关参数中,提取主要特征;使用局部加权回归对数据进行过滤处理,再结合LSTM建立下渣预测模型;给出LSTM模型与ARIMA模型、RNN模型的预测结果比较。研究结果表明,长短时记忆神经网络模型的预测误差小,预测准确度较高,具有广泛的应用前景。
For the problem of precise control of the slag time in the continuous casting process of ladle,a prediction model for continuous casting slag based on local weighted regression and the Long Short-Term Memory(LSTM)neural network is proposed.The system processes and identifies the signals collected during the slag process,and can accurately predict the slag time.Combined with the actual production situation of a steel mill,the main features are extracted from the relevant parameters of the large amount of ladle collected.The data is filtered by local weighted regression and then the slag prediction model is established by combining LSTM.Then the comparison results of the LSTM model with the ARIMA model and the RNN model are given.The research results show that the LSTM-based neural network has small prediction error and high prediction accuracy and has broad application prospects.
作者
李福进
刘尚瑜
史涛
LI Fu-jin;LIU Shang-yu;SHI Tao(College of Electrical Engineering,North China University of Science and Technology,Hebei Tangshan 063210,China)
出处
《机械设计与制造》
北大核心
2022年第1期181-183,188,共4页
Machinery Design & Manufacture
基金
河北省自然科学基金资助项目(F2018209289)。
关键词
钢包下渣
递归神经网络
长短时记忆
局部加权回归
时间序列预测
Ladle Slag
Recurrent Neural Network
Long Short-Term Memory Neural Network
Local Weighted Regression
Time Series Prediction
