基于时间卷积网络的废钢价格预测模型  

Prediction Model on Steel Scrap Prices Based on Temporal Convolution Networks

作　　者：张洪 Zhang Hong(Industrial Internet Research Institute of Shanghai Baosight Software Co., Ltd./ Big Data Center, Shanghai 201203)

机构地区：[1]上海宝信软件股份有限公司工业互联网研究院/大数据中心,上海201203

出　　处：《武汉工程职业技术学院学报》2021年第2期26-30,共5页Journal of Wuhan Engineering Institute

摘　　要：针对废钢价格预测问题,引入了深度学习领域的时间卷积网络构建模型。通过数据预处理、特征工程、构建时间卷积层、构建深度神经网络等步骤,拟合了一个根据历史一个月的废钢价格数据推测未来一天废钢价格指数的函数。通过近期四个月的实际数据验证,在预测误差绝对值和趋势预测准确度两个指标上,均优于传统时序以及神经网络模型。深度学习在废钢价格预测方面的研究也为其应用于行业其他时序相关数据的分析和建模提供了参考方向。Application of temporal convolution networks in the deep learning field aiming at prediction of steel scrap prices is studied in this research.Through processes of data preprocessing,feature engineering,building-up of temporal convolution layers and construction of deep neural networks,a model is designed to predict the steel scrap price in the next day according to scrap price data of one month in history.Results of a 4-month experiment prove the validity of the model,outperforming most of traditional models for financial time series prediction with less absolute prediction error and better trend prediction accuracy.The research of deep learning in scrap price prediction also provides a reference direction for its application in the analysis and modeling of other time-series related data in the industry.

关 键 词：废钢 价格预测 深度学习 时间卷积网络 卷积神经网络 循环神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]