摘要
【目的】中国作为全球最大的天然气进口国,LNG在天然气供给中扮演着关键角色。然而中国LNG出厂价格的变化呈现较强的不确定性与波动性,准确预测中国LNG出厂价格的变化趋势,对LNG产业链布局优化、提高天然气供应链的经济性具有重要现实意义。【方法】研究收集并整理了中国LNG出厂价格的历史数据,并采用灰色关联分析确定了影响价格变动的关键因素。在此基础上,分别构建BP神经网络预测模型与ARIMA时间序列预测模型。根据变权重理论,将两种单一预测模型进行加权组合,建立新的变权重BP-ARIMA组合预测模型。以中国实际LNG出厂价格数据对不同预测模型进行案例验证分析。【结果】与传统的BP神经网络模型及ARIMA模型相比,变权重BP-ARIMA组合模型结合了两种单一模型的优势,通过动态调整两种模型的权重,显著提高了LNG出厂价格预测的准确性。其平均绝对误差MAE、平均绝对百分比误差MAPE以及均方根误差RMSE分别为188.7元/t、4.1%、280.5元/t,相较于BP神经网络模型、ARIMA模型以及等权重组合模型,MAE分别降低了65.85%、44.20%、37.50%,RMSE降低了63.6%、38.2%、29.8%,MAPE降低了63.7%、42.3%、36.9%。【结论】变权重BP-ARIMA组合预测模型为中国LNG出厂价格预测难题提供了有效解决方案,有助于促进LNG市场有序发展并为天然气供应链管理提供决策支持。
[Objective]China is the world’s largest importer of natural gas,with LNG playing a crucial role in natural gas supply.However,fluctuations in LNG ex-factory prices introduce significant uncertainty.Accurately predicting these price trends is vital for optimizing the LNG industry chain layout and enhancing the economic efficiency of the natural gas supply chain.[Methods]Historical data on China’s LNG ex-factory prices were collected and analyzed,identifying key factors that influence price changes through grey relational analysis.On this basis,BP neural network prediction model and ARIMA time series prediction model were established,respectively.According to the variable weight theory,a new variable-weight BP-ARIMA combined prediction model was established through the weighted combination of the two individual prediction models.Case verification analysis was carried out for different prediction models based on China’s actual LNG ex-factory prices.[Results]The variable-weight BP-ARIMA combined model integrated the advantages of the traditional BP neural network and ARIMA models.By dynamically adjusting the weight ratio between the two,it significantly enhanced the accuracy of LNG ex-factory price predictions.The mean absolute error(MAE),mean absolute percentage error(MAPE),and root-mean-square error(RMSE)were RMB 188.7/t,4.1%,and RMB 280.5/t,respectively.Compared with the BP neural network model,ARIMA model,and equal-weight combined model,the MAE was reduced by 65.85%,44.20%and 37.50%,the RMSE by 63.6%,38.2%and 29.8%,and the MAPE by 63.7%,42.3%and 36.9%,respectively.[Conclusion]The variable-weight BP-ARIMA combined prediction model proposed in this study offers an effective solution for predicting China’s LNG ex-factory prices,facilitating the orderly development of the LNG market and providing support for decision-making in natural gas supply chain management.
作者
杜学平
赵清华
秘琳
郎智凯
柳梦琳
吴江涛
DU Xueping;ZHAO Qinghua;MI Lin;LANG Zhikai;LIU Menglin;WU Jiangtao(School of Energy and Power Engineering,Xi'an Jiaotong University;Beijing Enterprises Group Company Limited;Beijing Gas Group Co.Ltd.)
出处
《油气储运》
CAS
北大核心
2024年第10期1173-1179,1188,共8页
Oil & Gas Storage and Transportation
