摘要
钢铁物流是衡量无水港钢铁贸易水平的核心指标之一。针对提升无水港运作效率,提前规划库位分配并监测钢卷吞吐量,提出一种基于时间序列和极端梯度提升(Xgboost)的钢卷仓储吞吐量预测方法。该方法对输入特征中含有连续变量和离散变量的混杂系统构建时间特征,用滑窗法构造回归数据,用one-hot编码处理离散变量,将差分整合移动平均自回归(ARIMA)模型和Xgboost模型加权融合,实现热卷和冷卷未来7天和4周的吞吐量预测。利用某无水港2014年至2018年的历史销售订单、仓储、吞吐量等数据进行训练和测试。实验结果表明,ARIMA和Xgboost的组合模型得分82.4215,具有最高的预测精度,比单一的时间序列模型和机器学习模型准确率都高。
Steel logistics is one of the core indicators for measuring the level of steel trade in dry ports. In order to improve the efficiency of the dry port operation, plan the location allocation and monitor the coil throughput in advance,a time series and eXtreme gradient boosting(Xgboost)-based steel coil storage throughput prediction method was proposed. Temporal features were built for hybrid systems with continuous and discrete variables in the input characteristics, regression data was constructed by sliding window method, discrete variables were processed by one-hot encoding, AutoRegressive Integrated Moving Average(ARIMA) model and Xgboost model were weightedly fused, thus achieving the throughput prediction for hot and cold volumes for the next 7 and 4 weeks.The historical sales orders, warehousing, throughput and other data of a dry port from 2014 to 2018 were used for training and testing.The experimental results show that the combination model of ARIMA and Xgboost with score of 82.421 5 has the highest prediction accuracy, compared to single time series model and machine learning model.
作者
孟杭
黄细霞
涂修建
MENG Hang;HUANG Xixia;TU Xiujian(Key Laboratory of Maritime Technology and Control Engineering of Ministry of Communications(Shanghai Maritime University),Shanghai 201306,China)
出处
《计算机应用》
CSCD
北大核心
2019年第S02期24-28,共5页
journal of Computer Applications
基金
国家自然科学基金资助项目(61304186)
关键词
编码处理
混杂系统
时间序列
离散变量
移动平均
无水港
钢卷
仓储
dry port
steel coil throughput prediction
hybrid system
time feature
one-hot encoding
AutoRegressive Integrated Moving Average(ARIMA)
eXtreme gradient boosting(Xgboost)
